WO2021041862A1 - Deep learning aided mmwave mimo blind detection schemes - Google Patents

Abstract

Some embodiments of a method may include: receiving a signal and performing analog RF combining; down-converting the combined received signal; providing the down-converted combined signal to an artificial neural network (ANN); using predetermined weights to obtain log likelihood ratios (LLRs) in the absence of channel state information; and processing the LLRs with a channel decoder to obtain uncoded bits.

Description

DEEP LEARNING AIDED MMWAVE MIMO BLIND DETECTION SCHEMES

CROSS-REFERENCE TO RELATED APPLICATIONS

[0001] The present application is a non-provisional filing of, and claims benefit under 35 U.S.C. § 119(e) from, U.S. Provisional Patent Application Serial No. 62/894,464, entitled “Deep Learning Aided mmWave MIMO Blind Detection Schemes” and filed August 30, 2019, which is hereby incorporated by reference in its entirety.

BACKGROUND

[0002] Owing to the large available bandwidth, mmWave frequencies have the potential to accommodate a large number of users while simultaneously providing high data rates. However, harnessing mmWave frequencies faces several technical challenges, because mmWave frequencies suffer from high propagation losses compared to that of the sub-6 GHz spectrum.

[0003] Furthermore, the use of multiple-input multiple-output (MIMO) transmission has been beneficial for the enhancement of the data rates. More explicitly, the concept of multi-functional (MF) MIMO has been proposed as the amalgamation of diversity, multiplexing and beamforming. Among other MF MIMO techniques, space-time shift keying (STSK) offers a design trade-off between multiplexing and diversity gains. The STSK design is conceived as an extension to the concept of spatial modulation, such that a single antenna is activated at any time. To elaborate a little further, in the STSK design, a single dispersion matrix (DM) is activated amongst a set of DMs at any time. Information is conveyed by the index of the DM in addition to the complex-valued signal drawn from the M-ary constellation. As an extension of the STSK, a multi-set (MS) STSK is formed by combining the concepts of the STSK and spatial modulation (SM). This design may increase the data rate because the information is carried by both the M-ary alphabet and DM index as well as the antenna index combination. In mmWave communications, if the channel supports a few clusters of rays, the data rate of the MS-STSK design may be further enhanced by coupling with the concept of beam index modulation (BIM). In BIM-aided transmission, information is conveyed by the index of the beams in addition to the M-ary constellation. SUMMARY

[0004] A further example method in accordance with some embodiments may include: receiving a signal and performing analog RF combining to generate a combined received signal; down-converting the combined received signal; providing the down-converted combined signal to an artificial neural network (ANN); and using predetermined weights to demodulate the received signal in absence of channel state information to obtain data bits.

[0005] Some embodiments of the further example method may further include recalibrating periodically the predetermined weights using training update information transmitted from the transmitter to the receiver.

[0006] Some embodiments of the further example method may further include storing at least one of the data bits or the received, combined, and down-converted signal of received signal values as input samples to a replay buffer.

[0007] For some embodiments of the further example method, the input samples may be stored in the replay buffer only if a cyclic redundancy check (CRC) passes.

[0008] Some embodiments of the further example method may further include storing a time stamp indicating at least one of when the input samples were collected and how long the input samples will stay in the replay buffer.

[0009] For some embodiments of the further example method, a packet of uncoded bits from the replay buffer may be re-encoded to reproduce a transmitter output and corresponding target output of the ANN.

[0010] For some embodiments of the further example method, at least one of: (i) an estimate of a channel coherence time, (ii) a recent bit error rate, or (iii) a packet error rate, is used to determine how long at least one replay buffer entry will remain in the replay buffer.

[0011] For some embodiments of the further example method, obtaining uncoded bits may include: inputting to the ANN a vector y, representing a vectorized matrix, comprising the down-converted combined signal of received signal values; and obtaining an ANN output vector comprising a detected dispersion matrix index, an antenna index, and a complex-valued symbol drawn from an M-QAM constellation.

[0012] For some embodiments of the further example method, the ANN may be configured to have 2 hidden layers.

[0013] Some embodiments of the further example method may further include adjusting one or more neurons in at least one level of the ANN to reproduce a target output during a training stage. [0014] For some embodiments of the further example method, the received signal may be modulated at a transmitter using MS-STSK symbol transmission.

[0015] For some embodiments of the further example method, the received signal may be modulated at a transmitter with an MS-STSK encoder in combination with beam index modulation, and the vectorized matrix y may be equal to a vectorized matrix y_BI corresponding to beam index modulation.

[0016] For some embodiments of the further example method, the ANN output vector may further include a beam index.

[0017] For some embodiments of the further example method, the ANN may include weights and biases determined during a training stage to reduce an error between a target output and a predicted output.

[0018] For some embodiments of the further example method, the ANN may include activating functions of a type of hyperbolic tangent function, or a type of sigmoid function.

[0019] For some embodiments of the further example method, the received signal may be divided into real and imaginary values by splitting the received vector y into a real part Real(y) and an imaginary part lmag(y) before providing the real part Real(y) as the input to the ANN.

[0020] For some embodiments of the further example method, training data for recalibrating the ANN weights may be sent after every N_f frames.

[0021] For some embodiments of the further example method, blind-detection may be performed for frames received between training frames.

[0022] Some embodiments of the further example method may further include storing the received signal as input samples to a replay buffer.

[0023] A further example apparatus in accordance with some embodiments may include: a processor; and a non-transitory computer-readable medium storing instructions operative, when executed by the processor, to: receive a signal and perform analog RF combining; down-convert the combined received signal; provide the down-converted combined signal to an artificial neural network (ANN); and demodulate the received signal using predetermined weights in absence of channel state information to obtain uncoded bits.

[0024] A further additional example method in accordance with some embodiments may include: using a vector y, representing a vectorized matrix, comprising a received signal of received signal values as an input to an artificial neural network (ANN); and obtaining an ANN output vector comprising a detected dispersion matrix index, an antenna index, and a complex-valued symbol drawn from an M-QAM constellation. [0025] For some embodiments of the further additional example method, the ANN may be configured to have 2 hidden layers.

[0026] Some embodiments of the further additional example method may further include adjusting one or more neurons in at least one level of the ANN to reproduce a target output during a training stage.

[0027] For some embodiments of the further additional example method, the received signal may be modulated at a transmitter using MS-STSK symbol transmission.

[0028] For some embodiments of the further additional example method, the received signal may be modulated at a transmitter with an MS-STSK encoder in combination with beam-index modulation, and a vectorized matrix y_BI may be used as the input to the ANN.

[0029] For some embodiments of the further additional example method, the ANN output vector may further include a beam index.

[0030] For some embodiments of the further additional example method, the ANN may include weights and biases determined during a training stage to reduce an error between a target output and a predicted output.

[0031] For some embodiments of the further additional example method, the ANN may include activating functions of a type of hyperbolic tangent function, or a type of sigmoid function.

[0032] For some embodiments of the further additional example method, the received signal may be divided into real and imaginary values by splitting the received vector y into a real part Real(y) and an imaginary part lmag(y) before providing the real part Real(y) as the input to the ANN.

[0033] For some embodiments of the further additional example method, training data for recalibrating the ANN weights may be sent after every N_f frames.

[0034] For some embodiments of the further additional example method, blind-detection may be performed for frames received between training frames.

[0035] Some embodiments of the further additional example method may further include storing the received signal as input samples to a replay buffer.

[0036] A further additional example apparatus in accordance with some embodiments may include: a processor; and a non-transitory computer-readable medium storing instructions operative, when executed by the processor, to: use a vectorized matrix y comprising a received signal of received signal values as an input to an artificial neural network (ANN); and obtain an ANN output vector comprising a detected dispersion matrix index, an antenna index, and a complex-valued symbol drawn from an M-QAM constellation. [0037] Another example apparatus in accordance with some embodiments may include: an RF combiner; a downconverter; an artificial neural network (ANN) module; and a forward error correction (FEC) decoder;

[0038] Some embodiments of another example apparatus may further include a replay buffer.

[0039] Another further example method in accordance with some embodiments may include: establishing a beam alignment between a base station and a user equipment; receiving training data from the base station; training a neural network to detect a data signal from the base station using the training data; receiving a real data transmission signal from the base station; checking a key performance indicator (KPI) with a pre-determined value; and requesting retraining of the neural network if the KPI is less than the predetermined value.

[0040] Another further example apparatus in accordance with some embodiments may include: a processor; and a non-transitory computer-readable medium storing instructions operative, when executed by the processor, to: establish a beam alignment between a base station and a user equipment; receive training data from the base station; train a neural network to detect a data signal from the base station using the training data; receive a real data transmission signal from the base station; check a key performance indicator (KPI) with a pre-determined value; and request retraining of the neural network if the KPI is less than the pre-determined value.

BRIEF DESCRIPTION OF THE DRAWINGS

[0041] FIG. 1A is a system diagram illustrating an example communications system according to some embodiments.

[0042] FIG. 1 B is a system diagram illustrating an example wireless transmit/receive unit (WTRU) that may be used within the communications system illustrated in FIG. 1A according to some embodiments.

[0043] FIG. 1C is a system diagram of an example system illustrating an example radio access network (RAN) and an example core network (CN) that may be used within the communications system illustrated in FIG. 1A according to some embodiments.

[0044] FIG. 1 D is a system diagram of an example system illustrating a further example RAN and a further example CN that may be used within the communications system illustrated in FIG. 1A according to some embodiments.

[0045] FIG. 2 is a schematic illustration showing an example MS-STSK encoder according to some embodiments. [0046] FIG. 3 is a schematic illustration showing an example beamformed MS-STSK symbol transmitted in the direction of a desired user according to some embodiments.

[0047] FIG. 4 is a schematic illustration showing an example beamformed MS-STSK symbol coupled with the beam index transmitted in the direction of a desired user according to some embodiments.

[0048] FIG. 5 is a schematic illustration showing an example neural network model according to some embodiments.

[0049] FIG. 6A is a schematic illustration showing a first example of ANN assisted blind detection at the receiver according to some embodiments.

[0050] FIG. 6B is a schematic illustration showing a second example of ANN assisted blind detection at the receiver according to some embodiments.

[0051] FIG. 7 is a functional block diagram illustrating an example of conventional maximum-likelihood detection for MS-STSK transmissions.

[0052] FIG. 8 is a functional block diagram illustrating an example of learning assisted detection for MS- STSK transmissions according to some embodiments.

[0053] FIG. 9 is a message sequencing diagram illustrating an example process of deep learning detection for MS-STSK transmissions according to some embodiments.

[0054] FIG. 10 is a message sequencing diagram illustrating an example process of deep learning detection for MS-STSK transmissions with beam index modulation according to some embodiments.

[0055] FIG. 11 is a graph illustrating example bit error rates for MS-STSK transmissions according to some embodiments.

[0056] FIG. 12 is a graph illustrating example discrete-input continuous-output memoryless channel capacities for MS-STSK transmissions according to some embodiments.

[0057] FIG. 13 is a graph illustrating example bit error rates for MS-STSK transmissions with beam index modulation according to some embodiments.

[0058] FIG. 14 is a graph illustrating example discrete-input continuous-output memoryless channel capacities for MS-STSK transmissions with beam index modulation according to some embodiments.

[0059] FIG. 15 is a graph illustrating example bit error rates for different numbers of frames and Jake’s correlation coefficient values according to some embodiments. [0060] FIG. 16A is a frame structure diagram illustrating an example maximum-likelihood (ML)-assisted detection according to some embodiments.

[0061] FIG. 16B is a frame structure diagram illustrating an example learning-assisted detection according to some embodiments.

[0062] FIG. 17 is a schematic illustration showing example inputs and outputs of the ANN for soft detection according to some embodiments.

[0063] FIG. 18 is a functional block diagram illustrating an example ML assisted detection for MS-STSK transmissions.

[0064] FIG. 19 is a functional block diagram illustrating an example deep learning assisted detection for MS-STSK transmissions according to some embodiments.

[0065] FIG. 20 is a graph illustrating example bit error rates for conventional and learning aided soft demodulation according to some embodiments.

[0066] FIG. 21 is a graph illustrating an example discrete-input continuous-output memoryless channel capacities for conventional and learning aided soft demodulation for 3%, 5%, and 10% pilot overheads according to some embodiments.

[0067] FIG. 22 is a flowchart illustrating an example process for obtaining uncoded bits using an ANN according to some embodiments.

[0068] FIG. 23 is a flowchart illustrating an example process for obtaining an ANN output vector according to some embodiments.

[0069] The entities, connections, arrangements, and the like that are depicted in— and described in connection with— the various figures are presented by way of example and not by way of limitation. As such, any and all statements or other indications as to what a particular figure “depicts,” what a particular element or entity in a particular figure “is” or “has,” and any and all similar statements— that may in isolation and out of context be read as absolute and therefore limiting— may only properly be read as being constructively preceded by a clause such as “In at least one embodiment, ....” For brevity and clarity of presentation, this implied leading clause is not repeated ad nauseum in the detailed description.

EXAMPLE NETWORKS FOR IMPLEMENTATION OF THE EMBODIMENTS

[0070] A wireless transmit/receive unit (WTRU) may be used, e.g., as a handset, smartphone, or mobile device (which may be indicated as a user) in some embodiments described herein. [0071] FIG. 1A is a diagram illustrating an example communications system 100 in which one or more disclosed embodiments may be implemented. The communications system 100 may be a multiple access system that provides content, such as voice, data, video, messaging, broadcast, etc., to multiple wireless users. The communications system 100 may enable multiple wireless users to access such content through the sharing of system resources, including wireless bandwidth. For example, the communications systems 100 may employ one or more channel access methods, such as code division multiple access (CDMA), time division multiple access (TDMA), frequency division multiple access (FDMA), orthogonal FDMA (OFDMA), single-carrier FDMA (SC-FDMA), zero-tail unique-word DFT-Spread OFDM (ZT UW DTS-s OFDM), unique word OFDM (UW-OFDM), resource block-filtered OFDM, filter bank multicarrier (FBMC), and the like.

[0072] As shown in FIG. 1A, the communications system 100 may include wireless transmit/receive units (WTRUs) 102a, 102b, 102c, 102d, a RAN 104/113, a ON 106, a public switched telephone network (PSTN) 108, the Internet 110, and other networks 112, though it will be appreciated that the disclosed embodiments contemplate any number of WTRUs, base stations, networks, and/or network elements. Each of the WTRUs 102a, 102b, 102c, 102d may be any type of device configured to operate and/or communicate in a wireless environment. By way of example, the WTRUs 102a, 102b, 102c, 102d, any of which may be referred to as a “station” and/or a “STA”, may be configured to transmit and/or receive wireless signals and may include a user equipment (UE), a mobile station, a fixed or mobile subscriber unit, a subscription-based unit, a pager, a cellular telephone, a personal digital assistant (PDA), a smartphone, a laptop, a netbook, a personal computer, a wireless sensor, a hotspot or Mi-Fi device, an Internet of Things (loT) device, a watch or other wearable, a head-mounted display (HMD), a vehicle, a drone, a medical device and applications (e.g., remote surgery), an industrial device and applications (e.g., a robot and/or other wireless devices operating in an industrial and/or an automated processing chain contexts), a consumer electronics device, a device operating on commercial and/or industrial wireless networks, and the like. Any of the WTRUs 102a, 102b, 102c and 102d may be interchangeably referred to as a UE.

[0073] The communications systems 100 may also include a base station 114a and/or a base station 114b. Each of the base stations 114a, 114b may be any type of device configured to wirelessly interface with at least one of the WTRUs 102a, 102b, 102c, 102d to facilitate access to one or more communication networks, such as the CN 106, the Internet 110, and/or the other networks 112. By way of example, the base stations 114a, 114b may be a base transceiver station (BTS), a Node-B, an eNode B, a Flome Node B, a Flome eNode B, a gNB, a NR NodeB, a site controller, an access point (AP), a wireless router, and the like. While the base stations 114a, 114b are each depicted as a single element, it will be appreciated that the base stations 114a, 114b may include any number of interconnected base stations and/or network elements. [0074] The base station 114a may be part of the RAN 104/113, which may also include other base stations and/or network elements (not shown), such as a base station controller (BSC), a radio network controller (RNC), relay nodes, etc. The base station 114a and/or the base station 114b may be configured to transmit and/or receive wireless signals on one or more carrier frequencies, which may be referred to as a cell (not shown). These frequencies may be in licensed spectrum, unlicensed spectrum, or a combination of licensed and unlicensed spectrum. A cell may provide coverage for a wireless service to a specific geographical area that may be relatively fixed or that may change over time. The cell may further be divided into cell sectors. For example, the cell associated with the base station 114a may be divided into three sectors. Thus, in one embodiment, the base station 114a may include three transceivers, i.e., one for each sector of the cell. In an embodiment, the base station 114a may employ multiple-input multiple output (MIMO) technology and may utilize multiple transceivers for each sector of the cell. For example, beamforming may be used to transmit and/or receive signals in desired spatial directions.

[0075] The base stations 114a, 114b may communicate with one or more of the WTRUs 102a, 102b, 102c, 102d over an air interface 116, which may be any suitable wireless communication link (e.g., radio frequency (RF), microwave, centimeter wave, micrometer wave, infrared (IR), ultraviolet (UV), visible light, etc.). The air interface 116 may be established using any suitable radio access technology (RAT).

[0076] More specifically, as noted above, the communications system 100 may be a multiple access system and may employ one or more channel access schemes, such as CDMA, TDMA, FDMA, OFDMA, SC-FDMA, and the like. For example, the base station 114a in the RAN 104/113 and the WTRUs 102a, 102b, 102c may implement a radio technology such as Universal Mobile Telecommunications System (UMTS) Terrestrial Radio Access (UTRA), which may establish the air interface 116 using wideband CDMA (WCDMA). WCDMA may include communication protocols such as High-Speed Packet Access (HSPA) and/or Evolved HSPA (FISPA+). HSPA may include High-Speed Downlink (DL) Packet Access (FISDPA) and/or High-Speed UL Packet Access (FISUPA).

[0077] In an embodiment, the base station 114a and the WTRUs 102a, 102b, 102c may implement a radio technology such as Evolved UMTS Terrestrial Radio Access (E-UTRA), which may establish the air interface 116 using Long Term Evolution (LTE) and/or LTE-Advanced (LTE-A) and/or LTE-Advanced Pro (LTE-A Pro).

[0078] In an embodiment, the base station 114a and the WTRUs 102a, 102b, 102c may implement a radio technology such as NR Radio Access , which may establish the air interface 116 using New Radio (NR). 10079] In an embodiment, the base station 114a and the WTRUs 102a, 102b, 102c may implement multiple radio access technologies. For example, the base station 114a and the WTRUs 102a, 102b, 102c may implement LTE radio access and NR radio access together, for instance using dual connectivity (DC) principles. Thus, the air interface utilized by WTRUs 102a, 102b, 102c may be characterized by multiple types of radio access technologies and/or transmissions sent to/from multiple types of base stations (e.g., a eNB and a gNB).

[0080] In other embodiments, the base station 114a and the WTRUs 102a, 102b, 102c may implement radio technologies such as IEEE 802.11 (i.e., Wireless Fidelity (WiFi), IEEE 802.16 (i.e., Worldwide Interoperability for Microwave Access (WiMAX)), CDMA2000, CDMA2000 1X, CDMA2000 EV-DO, Interim Standard 2000 (IS-2000), Interim Standard 95 (IS-95), Interim Standard 856 (IS-856), Global System for Mobile communications (GSM), Enhanced Data rates for GSM Evolution (EDGE), GSM EDGE (GERAN), and the like.

[0081] The base station 114b in FIG. 1A may be a wireless router, Flome Node B, Flome eNode B, or access point, for example, and may utilize any suitable RAT for facilitating wireless connectivity in a localized area, such as a place of business, a home, a vehicle, a campus, an industrial facility, an air corridor (e.g., for use by drones), a roadway, and the like. In one embodiment, the base station 114b and the WTRUs 102c, 102d may implement a radio technology such as IEEE 802.11 to establish a wireless local area network (WLAN). In an embodiment, the base station 114b and the WTRUs 102c, 102d may implement a radio technology such as IEEE 802.15 to establish a wireless personal area network (WPAN). In yet another embodiment, the base station 114b and the WTRUs 102c, 102d may utilize a cellular-based RAT (e.g., WCDMA, CDMA2000, GSM, LTE, LTE-A, LTE-A Pro, NR etc.) to establish a picocell orfemtocell. As shown in FIG. 1A, the base station 114b may have a direct connection to the Internet 110. Thus, the base station 114b may not be required to access the Internet 110 via the CN 106.

[0082] The RAN 104/113 may be in communication with the CN 106, which may be any type of network configured to provide voice, data, applications, and/or voice over internet protocol (VoIP) services to one or more of the WTRUs 102a, 102b, 102c, 102d. The data may have varying quality of service (QoS) requirements, such as differing throughput requirements, latency requirements, error tolerance requirements, reliability requirements, data throughput requirements, mobility requirements, and the like. The CN 106 may provide call control, billing services, mobile location-based services, pre-paid calling, Internet connectivity, video distribution, etc., and/or perform high-level security functions, such as user authentication. Although not shown in FIG. 1A, it will be appreciated that the RAN 104/113 and/or the CN 106 may be in direct or indirect communication with other RANs that employ the same RAT as the RAN 104/113 or a different RAT. For example, in addition to being connected to the RAN 104/113, which may be utilizing a NR radio technology, the CN 106 may also be in communication with another RAN (not shown) employing a GSM, UMTS, CDMA 2000, WiMAX, E-UTRA, or WiFi radio technology.

[0083] The CN 106 may also serve as a gateway for the WTRUs 102a, 102b, 102c, 102d to access the PSTN 108, the Internet 110, and/or the other networks 112. The PSTN 108 may include circuit-switched telephone networks that provide plain old telephone service (POTS). The Internet 110 may include a global system of interconnected computer networks and devices that use common communication protocols, such as the transmission control protocol (TCP), user datagram protocol (UDP) and/or the internet protocol (IP) in the TCP/IP internet protocol suite. The networks 112 may include wired and/or wireless communications networks owned and/or operated by other service providers. For example, the networks 112 may include another CN connected to one or more RANs, which may employ the same RAT as the RAN 104/113 or a different RAT.

[0084] Some or all of the WTRUs 102a, 102b, 102c, 102d in the communications system 100 may include multi-mode capabilities (e.g., the WTRUs 102a, 102b, 102c, 102d may include multiple transceivers for communicating with different wireless networks over different wireless links). For example, the WTRU 102c shown in FIG. 1A may be configured to communicate with the base station 114a, which may employ a cellular-based radio technology, and with the base station 114b, which may employ an IEEE 802 radio technology.

[0085] FIG. 1 B is a system diagram illustrating an example WTRU 102. As shown in FIG. 1 B, the WTRU 102 may include a processor 118, a transceiver 120, a transmit/receive element 122, a speaker/microphone 124, a keypad 126, a display/touchpad 128, non-removable memory 130, removable memory 132, a power source 134, a global positioning system (GPS) chipset 136, and/or other peripherals 138, among others. It will be appreciated that the WTRU 102 may include any sub-combination of the foregoing elements while remaining consistent with an embodiment.

[0086] The processor 118 may be a general purpose processor, a special purpose processor, a conventional processor, a digital signal processor (DSP), a plurality of microprocessors, one or more microprocessors in association with a DSP core, a controller, a microcontroller, Application Specific Integrated Circuits (ASICs), Field Programmable Gate Arrays (FPGAs) circuits, any other type of integrated circuit (IC), a state machine, and the like. The processor 118 may perform signal coding, data processing, power control, input/output processing, and/or any other functionality that enables the WTRU 102 to operate in a wireless environment. The processor 118 may be coupled to the transceiver 120, which may be coupled to the transmit/receive element 122. While FIG. 1 B depicts the processor 118 and the transceiver 120 as separate components, it will be appreciated that the processor 118 and the transceiver 120 may be integrated together in an electronic package or chip.

[0087] The transmit/receive element 122 may be configured to transmit signals to, or receive signals from, a base station (e.g., the base station 114a) over the air interface 116. For example, in one embodiment, the transmit/receive element 122 may be an antenna configured to transmit and/or receive RF signals. In an embodiment, the transmit/receive element 122 may be an emitter/detector configured to transmit and/or receive IR, UV, or visible light signals, for example. In yet another embodiment, the transmit/receive element 122 may be configured to transmit and/or receive both RF and light signals. It will be appreciated that the transmit/receive element 122 may be configured to transmit and/or receive any combination of wireless signals.

[0088] Although the transmit/receive element 122 is depicted in FIG. 1 B as a single element, the WTRU 102 may include any number of transmit/receive elements 122. More specifically, the WTRU 102 may employ MIMO technology. Thus, in one embodiment, the WTRU 102 may include two or more transmit/receive elements 122 (e.g., multiple antennas) for transmitting and receiving wireless signals over the air interface 116.

[0089] The transceiver 120 may be configured to modulate the signals that are to be transmitted by the transmit/receive element 122 and to demodulate the signals that are received by the transmit/receive element 122. As noted above, the WTRU 102 may have multi-mode capabilities. Thus, the transceiver 120 may include multiple transceivers for enabling the WTRU 102 to communicate via multiple RATs, such as NR and IEEE 802.11, for example.

[0090] The processor 118 of the WTRU 102 may be coupled to, and may receive user input data from, the speaker/microphone 124, the keypad 126, and/or the display/touchpad 128 (e.g., a liquid crystal display (LCD) display unit or organic light-emitting diode (OLED) display unit). The processor 118 may also output user data to the speaker/microphone 124, the keypad 126, and/or the display/touchpad 128. In addition, the processor 118 may access information from, and store data in, any type of suitable memory, such as the non-removable memory 130 and/or the removable memory 132. The non-removable memory 130 may include random-access memory (RAM), read-only memory (ROM), a hard disk, or any other type of memory storage device. The removable memory 132 may include a subscriber identity module (SIM) card, a memory stick, a secure digital (SD) memory card, and the like. In other embodiments, the processor 118 may access information from, and store data in, memory that is not physically located on the WTRU 102, such as on a server or a home computer (not shown). [0091] The processor 118 may receive power from the power source 134, and may be configured to distribute and/or control the power to the other components in the WTRU 102. The power source 134 may be any suitable device for powering the WTRU 102. For example, the power source 134 may include one or more dry cell batteries (e.g., nickel-cadmium (NiCd), nickel-zinc (NiZn), nickel metal hydride (NiMH), lithium- ion (Li-ion), etc.), solar cells, fuel cells, and the like.

[0092] The processor 118 may also be coupled to the GPS chipset 136, which may be configured to provide location information (e.g., longitude and latitude) regarding the current location of the WTRU 102. In addition to, or in lieu of, the information from the GPS chipset 136, the WTRU 102 may receive location information over the air interface 116 from a base station (e.g., base stations 114a, 114b) and/or determine its location based on the timing of the signals being received from two or more nearby base stations. It will be appreciated that the WTRU 102 may acquire location information by way of any suitable location- determination method while remaining consistent with an embodiment.

[0093] The processor 118 may further be coupled to other peripherals 138, which may include one or more software and/or hardware modules that provide additional features, functionality and/or wired or wireless connectivity. For example, the peripherals 138 may include an accelerometer, an e-compass, a satellite transceiver, a digital camera (for photographs and/or video), a universal serial bus (USB) port, a vibration device, a television transceiver, a hands free headset, a Bluetooth® module, a frequency modulated (FM) radio unit, a digital music player, a media player, a video game player module, an Internet browser, a Virtual Reality and/or Augmented Reality (VR/AR) device, an activity tracker, and the like. The peripherals 138 may include one or more sensors, the sensors may be one or more of a gyroscope, an accelerometer, a hall effect sensor, a magnetometer, an orientation sensor, a proximity sensor, a temperature sensor, a time sensor; a geolocation sensor; an altimeter, a light sensor, a touch sensor, a magnetometer, a barometer, a gesture sensor, a biometric sensor, and/or a humidity sensor.

[0094] The WTRU 102 may include a full duplex radio for which transmission and reception of some or all of the signals (e.g., associated with particular subframes for both the UL (e.g., for transmission) and downlink (e.g., for reception) may be concurrent and/or simultaneous. The full duplex radio may include an interference management unit to reduce and or substantially eliminate self-interference via either hardware (e.g., a choke) or signal processing via a processor (e.g., a separate processor (not shown) or via processor 118). In an embodiment, the WTRU 102 may include a half-duplex radio for which transmission and reception of some or all of the signals (e.g., associated with particular subframes for either the UL (e.g., for transmission) or the downlink (e.g., for reception)). [0095] FIG. 1 C is a system diagram illustrating the RAN 104 and the CN 106 according to an embodiment. As noted above, the RAN 104 may employ an E-UTRA radio technology to communicate with the WTRUs 102a, 102b, 102c over the air interface 116. The RAN 104 may also be in communication with the CN 106.

[0096] The RAN 104 may include eNode-Bs 160a, 160b, 160c, though it will be appreciated that the RAN 104 may include any number of eNode-Bs while remaining consistent with an embodiment. The eNode-Bs 160a, 160b, 160c may each include one or more transceivers for communicating with the WTRUs 102a, 102b, 102c over the air interface 116. In one embodiment, the eNode-Bs 160a, 160b, 160c may implement MIMO technology. Thus, the eNode-B 160a, for example, may use multiple antennas to transmit wireless signals to, and/or receive wireless signals from, the WTRU 102a.

[0097] Each of the eNode-Bs 160a, 160b, 160c may be associated with a particular cell (not shown) and may be configured to handle radio resource management decisions, handover decisions, scheduling of users in the UL and/or DL, and the like. As shown in FIG. 1C, the eNode-Bs 160a, 160b, 160c may communicate with one another over an X2 interface.

[0098] The CN 106 shown in FIG. 1C may include a mobility management entity (MME) 162, a serving gateway (SGW) 164, and a packet data network (PDN) gateway (or PGW) 166. While each of the foregoing elements are depicted as part of the CN 106, it will be appreciated that any of these elements may be owned and/or operated by an entity other than the CN operator.

[0099] The MME 162 may be connected to each of the eNode-Bs 162a, 162b, 162c in the RAN 104 via an S1 interface and may serve as a control node. For example, the MME 162 may be responsible for authenticating users of the WTRUs 102a, 102b, 102c, bearer activation/deactivation, selecting a particular serving gateway during an initial attach of the WTRUs 102a, 102b, 102c, and the like. The MME 162 may provide a control plane function for switching between the RAN 104 and other RANs (not shown) that employ other radio technologies, such as GSM and/or WCDMA.

[0100] The SGW 164 may be connected to each of the eNode Bs 160a, 160b, 160c in the RAN 104 via the S1 interface. The SGW 164 may generally route and forward user data packets to/from the WTRUs 102a, 102b, 102c. The SGW 164 may perform other functions, such as anchoring user planes during inter-eNode B handovers, triggering paging when DL data is available for the WTRUs 102a, 102b, 102c, managing and storing contexts of the WTRUs 102a, 102b, 102c, and the like.

[0101] The SGW 164 may be connected to the PGW 166, which may provide the WTRUs 102a, 102b, 102c with access to packet-switched networks, such as the Internet 110, to facilitate communications between the WTRUs 102a, 102b, 102c and IP-enabled devices. [0102] The CN 106 may facilitate communications with other networks. For example, the CN 106 may provide the WTRUs 102a, 102b, 102c with access to circuit-switched networks, such as the PSTN 108, to facilitate communications between the WTRUs 102a, 102b, 102c and traditional land-line communications devices. For example, the CN 106 may include, or may communicate with, an IP gateway (e.g., an IP multimedia subsystem (IMS) server) that serves as an interface between the CN 106 and the PSTN 108. In addition, the CN 106 may provide the WTRUs 102a, 102b, 102c with access to the other networks 112, which may include other wired and/or wireless networks that are owned and/or operated by other service providers.

[0103] Although the WTRU is described in FIGS. 1 A-1 D as a wireless terminal, it is contemplated that in certain representative embodiments that such a terminal may use (e.g., temporarily or permanently) wired communication interfaces with the communication network.

[0104] In representative embodiments, the other network 112 may be a WLAN.

[0105] A WLAN in Infrastructure Basic Service Set (BSS) mode may have an Access Point (AP) for the BSS and one or more stations (STAs) associated with the AP. The AP may have an access or an interface to a Distribution System (DS) or another type of wired/wireless network that carries traffic in to and/or out of the BSS. Traffic to STAs that originates from outside the BSS may arrive through the AP and may be delivered to the STAs. Traffic originating from STAs to destinations outside the BSS may be sent to the AP to be delivered to respective destinations. Traffic between STAs within the BSS may be sent through the AP, for example, where the source STA may send traffic to the AP and the AP may deliver the traffic to the destination STA. The traffic between STAs within a BSS may be considered and/or referred to as peer-to- peer traffic. The peer-to-peer traffic may be sent between (e.g., directly between) the source and destination STAs with a direct link setup (DLS). In certain representative embodiments, the DLS may use an 802.11e DLS or an 802.11 z tunneled DLS (TDLS). A WLAN using an Independent BSS (IBSS) mode may not have an AP, and the STAs (e.g., all of the STAs) within or using the IBSS may communicate directly with each other. The IBSS mode of communication may sometimes be referred to herein as an “ad-hoc” mode of communication.

[0106] When using the 802.11 ac infrastructure mode of operation or a similar mode of operations, the AP may transmit a beacon on a fixed channel, such as a primary channel. The primary channel may be a fixed width (e.g., 20 MHz wide bandwidth) or a dynamically set width via signaling. The primary channel may be the operating channel of the BSS and may be used by the STAs to establish a connection with the AP. In certain representative embodiments, Carrier Sense Multiple Access with Collision Avoidance (CSMA/CA) may be implemented, for example in in 802.11 systems. For CSMA/CA, the STAs (e.g., every STA), including the AP, may sense the primary channel. If the primary channel is sensed/detected and/or determined to be busy by a particular STA, the particular STA may back off. One STA (e.g., only one station) may transmit at any given time in a given BSS.

[0107] High Throughput (HT) STAs may use a 40 MHz wide channel for communication, for example, via a combination of the primary 20 MHz channel with an adjacent or nonadjacent 20 MHz channel to form a 40 MHz wide channel.

[0108] Very High Throughput (VHT) STAs may support 20MHz, 40 MHz, 80 MHz, and/or 160 MHz wide channels. The 40 MHz, and/or 80 MHz, channels may be formed by combining contiguous 20 MHz channels. A 160 MHz channel may be formed by combining 8 contiguous 20 MHz channels, or by combining two noncontiguous 80 MHz channels, which may be referred to as an 80-^0 configuration. For the 80-^0 configuration, the data, after channel encoding, may be passed through a segment parser that may divide the data into two streams. Inverse Fast Fourier Transform (IFFT) processing, and time domain processing, may be done on each stream separately. The streams may be mapped on to the two 80 MHz channels, and the data may be transmitted by a transmitting STA. At the receiver of the receiving STA, the above described operation for the 80+80 configuration may be reversed, and the combined data may be sent to the Medium Access Control (MAC).

[0109] Sub 1 GHz modes of operation are supported by 802.11 af and 802.11 ah. The channel operating bandwidths, and carriers, are reduced in 802.11 af and 802.11 ah relative to those used in 802.11h, and 802.11ac. 802.11af supports 5 MHz, 10 MHz and 20 MHz bandwidths in the TV White Space (TVWS) spectrum, and 802.11 ah supports 1 MHz, 2 MHz, 4 MHz, 8 MHz, and 16 MHz bandwidths using non-TVWS spectrum. According to a representative embodiment, 802.11 ah may support Meter Type Control/Machine- Type Communications, such as MTC devices in a macro coverage area. MTC devices may have certain capabilities, for example, limited capabilities including support for (e.g., only support for) certain and/or limited bandwidths. The MTC devices may include a battery with a battery life above a threshold (e.g., to maintain a very long battery life).

[0110] WLAN systems, which may support multiple channels, and channel bandwidths, such as 802.11 n, 802.11ac, 802.11 af, and 802.11 ah, include a channel which may be designated as the primary channel. The primary channel may have a bandwidth equal to the largest common operating bandwidth supported by all STAs in the BSS. The bandwidth of the primary channel may be set and/or limited by a STA, from among all STAs in operating in a BSS, which supports the smallest bandwidth operating mode. In the example of 802.11 ah, the primary channel may be 1 MHz wide for STAs (e.g., MTC type devices) that support (e.g., only support) a 1 MHz mode, even if the AP, and other STAs in the BSS support 2 MHz, 4 MHz, 8 MHz, 16 MHz, and/or other channel bandwidth operating modes. Carrier sensing and/or Network Allocation Vector (NAV) settings may depend on the status of the primary channel. If the primary channel is busy, for example, due to a STA (which supports only a 1 MHz operating mode), transmitting to the AP, the entire available frequency bands may be considered busy even though a majority of the frequency bands remains idle and may be available.

[0111] In the United States, the available frequency bands, which may be used by 802.11 ah, are from 902 MHz to 928 MHz. In Korea, the available frequency bands are from 917.5 MHz to 923.5 MHz. In Japan, the available frequency bands are from 916.5 MHz to 927.5 MHz. The total bandwidth available for 802.11 ah is 6 MHz to 26 MHz depending on the country code.

[0112] FIG. 1 D is a system diagram illustrating the RAN 113 and the CN 115 according to an embodiment. As noted above, the RAN 113 may employ an NR radio technology to communicate with the WTRUs 102a, 102b, 102c over the air interface 116. The RAN 113 may also be in communication with the CN 115.

[0113] The RAN 113 may include gNBs 180a, 180b, 180c, though it will be appreciated that the RAN 113 may include any number of gNBs while remaining consistent with an embodiment. The gNBs 180a, 180b, 180c may each include one or more transceivers for communicating with the WTRUs 102a, 102b, 102c over the air interface 116. In one embodiment, the gNBs 180a, 180b, 180c may implement MIMO technology. For example, gNBs 180a, 108b may utilize beamforming to transmit signals to and/or receive signals from the gNBs 180a, 180b, 180c. Thus, the gNB 180a, for example, may use multiple antennas to transmit wireless signals to, and/or receive wireless signals from, the WTRU 102a. In an embodiment, the gNBs 180a, 180b, 180c may implement carrier aggregation technology. For example, the gNB 180a may transmit multiple component carriers to the WTRU 102a (not shown). A subset of these component carriers may be on unlicensed spectrum while the remaining component carriers may be on licensed spectrum. In an embodiment, the gNBs 180a, 180b, 180c may implement Coordinated Multi-Point (CoMP) technology. For example, WTRU 102a may receive coordinated transmissions from gNB 180a and gNB 180b (and/or gNB 180c).

[0114] The WTRUs 102a, 102b, 102c may communicate with gNBs 180a, 180b, 180c using transmissions associated with a scalable numerology. For example, the OFDM symbol spacing and/or OFDM subcarrier spacing may vary for different transmissions, different cells, and/or different portions of the wireless transmission spectrum. The WTRUs 102a, 102b, 102c may communicate with gNBs 180a, 180b, 180c using subframe or transmission time intervals (TTIs) of various or scalable lengths (e.g., containing varying number of OFDM symbols and/or lasting varying lengths of absolute time).

[0115] The gNBs 180a, 180b, 180c may be configured to communicate with the WTRUs 102a, 102b, 102c in a standalone configuration and/or a non-standalone configuration. In the standalone configuration, WTRUs 102a, 102b, 102c may communicate with gNBs 180a, 180b, 180c without also accessing other RANs (e.g., such as eNode-Bs 160a, 160b, 160c). In the standalone configuration, WTRUs 102a, 102b, 102c may utilize one or more of gNBs 180a, 180b, 180c as a mobility anchor point. In the standalone configuration, WTRUs 102a, 102b, 102c may communicate with gNBs 180a, 180b, 180c using signals in an unlicensed band. In a non-standalone configuration WTRUs 102a, 102b, 102c may communicate with/connect to gNBs 180a, 180b, 180c while also communicating with/connecting to another RAN such as eNode-Bs 160a, 160b, 160c. For example, WTRUs 102a, 102b, 102c may implement DC principles to communicate with one or more gNBs 180a, 180b, 180c and one or more eNode-Bs 160a, 160b, 160c substantially simultaneously. In the non- standalone configuration, eNode-Bs 160a, 160b, 160c may serve as a mobility anchor for WTRUs 102a, 102b, 102c and gNBs 180a, 180b, 180c may provide additional coverage and/or throughput for servicing WTRUs 102a, 102b, 102c.

[0116] Each of the gNBs 180a, 180b, 180c may be associated with a particular cell (not shown) and may be configured to handle radio resource management decisions, handover decisions, scheduling of users in the UL and/or DL, support of network slicing, dual connectivity, interworking between NR and E-UTRA, routing of user plane data towards User Plane Function (UPF) 184a, 184b, routing of control plane information towards Access and Mobility Management Function (AMF) 182a, 182b and the like. As shown in FIG. 1 D, the gNBs 180a, 180b, 180c may communicate with one another over an Xn interface.

[0117] The CN 115 shown in FIG. 1D may include at least one AMF 182a, 182b, at least one UPF 184a, 184b, at least one Session Management Function (SMF) 183a, 183b, and possibly a Data Network (DN) 185a, 185b. While each of the foregoing elements are depicted as part of the CN 115, it will be appreciated that any of these elements may be owned and/or operated by an entity other than the CN operator.

[0118] The AMF 182a, 182b may be connected to one or more of the gNBs 180a, 180b, 180c in the RAN 113 via an N2 interface and may serve as a control node. For example, the AMF 182a, 182b may be responsible for authenticating users of the WTRUs 102a, 102b, 102c, support for network slicing (e.g., handling of different PDU sessions with different requirements), selecting a particular SMF 183a, 183b, management of the registration area, termination of NAS signaling, mobility management, and the like. Network slicing may be used by the AMF 182a, 182b in order to customize CN support for WTRUs 102a, 102b, 102c based on the types of services being utilized WTRUs 102a, 102b, 102c. For example, different network slices may be established for different use cases such as services relying on ultra-reliable low latency (URLLC) access, services relying on enhanced massive mobile broadband (eMBB) access, services for machine type communication (MTC) access, and/or the like. The AMF 162 may provide a control plane function for switching between the RAN 113 and other RANs (not shown) that employ other radio technologies, such as LTE, LTE-A, LTE-A Pro, and/or non-3GPP access technologies such as WiFi.

[0119] The SMF 183a, 183b may be connected to an AMF 182a, 182b in the CN 115 via an N11 interface. The SMF 183a, 183b may also be connected to a UPF 184a, 184b in the CN 115 via an N4 interface. The SMF 183a, 183b may select and control the UPF 184a, 184b and configure the routing of traffic through the UPF 184a, 184b. The SMF 183a, 183b may perform other functions, such as managing and allocating UE IP address, managing PDU sessions, controlling policy enforcement and QoS, providing downlink data notifications, and the like. A PDU session type may be IP-based, non-IP based, Ethernet-based, and the like.

[0120] The UPF 184a, 184b may be connected to one or more of the gNBs 180a, 180b, 180c in the RAN 113 via an N3 interface, which may provide the WTRUs 102a, 102b, 102c with access to packet-switched networks, such as the Internet 110, to facilitate communications between the WTRUs 102a, 102b, 102c and IP-enabled devices. The UPF 184, 184b may perform other functions, such as routing and forwarding packets, enforcing user plane policies, supporting multi-homed PDU sessions, handling user plane QoS, buffering downlink packets, providing mobility anchoring, and the like.

[0121] The CN 115 may facilitate communications with other networks. For example, the CN 115 may include, or may communicate with, an IP gateway (e.g., an IP multimedia subsystem (IMS) server) that serves as an interface between the CN 115 and the PSTN 108. In addition, the CN 115 may provide the WTRUs 102a, 102b, 102c with access to the other networks 112, which may include other wired and/or wireless networks that are owned and/or operated by other service providers. In one embodiment, the WTRUs 102a, 102b, 102c may be connected to a local Data Network (DN) 185a, 185b through the UPF 184a, 184b via the N3 interface to the UPF 184a, 184b and an N6 interface between the UPF 184a, 184b and the DN 185a, 185b.

[0122] In view of Figures 1A-1D, and the corresponding description of Figures 1A-1 D, one or more, or all, of the functions described herein with regard to one or more of: WTRU 102a-d, Base Station 114a-b, eNode- B 160a-c, MME 162, SGW 164, PGW 166, gNB 180a-c, AMF 182a-b, UPF 184a-b, SMF 183a-b, DN 185a- b, and/or any other device(s) described herein, may be performed by one or more emulation devices (not shown). The emulation devices may be one or more devices configured to emulate one or more, or all, of the functions described herein. For example, the emulation devices may be used to test other devices and/or to simulate network and/or WTRU functions.

[0123] The emulation devices may be designed to implement one or more tests of other devices in a lab environment and/or in an operator network environment. For example, the one or more emulation devices may perform the one or more, or all, functions while being fully or partially implemented and/or deployed as part of a wired and/or wireless communication network in order to test other devices within the communication network. The one or more emulation devices may perform the one or more, or all, functions while being temporarily implemented/deployed as part of a wired and/or wireless communication network. The emulation device may be directly coupled to another device for purposes of testing and/or may performing testing using over-the-air wireless communications.

[0124] The one or more emulation devices may perform the one or more, including all, functions while not being implemented/deployed as part of a wired and/or wireless communication network. For example, the emulation devices may be utilized in a testing scenario in a testing laboratory and/or a non-deployed (e.g., testing) wired and/or wireless communication network in order to implement testing of one or more components. The one or more emulation devices may be test equipment. Direct RF coupling and/or wireless communications via RF circuitry (e.g., which may include one or more antennas) may be used by the emulation devices to transmit and/or receive data.

DETAILED DESCRIPTION

[0125] Systems and methods are described for a deep-learning aided semi-blind detection for mmWave MIMO using multi-dimensional index modulation, where the neural network (NN) weights are recalibrated after a certain number of frames, using side-information or training information transmitted from the transmitter to the receiver. Detection schemes are described for index modulation using the above semi-blind deep-learning design and not relying on the explicit knowledge of the channel state information (CSI) at the receiver, thus minimizing or avoiding the need for the pilot-assisted channel estimation. Soft-decoding schemes are described for multi-set space-time shift keying using the above semi-blind deep-learning design and not relying on the explicit knowledge of the CSI at the receiver, thus minimizing or avoiding the need for the pilot-assisted channel estimation. In some embodiments, experience replay buffers from received packets may be used to reduce or eliminate the need to send pilots for re-training the NN after the channel changes.

[0126] A growing concern in index modulation transmission schemes, such as the MS-STSK, is the search complexity at the receiver. Furthermore, like in any communication system, index modulation schemes impose the requirement of having accurate channel state information (CSI) at the receiver to achieve a low bit error rate (BER). In frequency division duplex (FDD) systems, the CSI estimate is typically carried out relying on pilots sent prior to the data transmission. Thus, this results in a reduced data rate in addition to the extra complexity involved for channel estimation before the final stage of detection is ensued. [0127] To circumvent this problem, a machine learning approach may be employed, where the symbol detection is carried out without explicit CSI knowledge. The receiver, relying on the learning strategy employed during the training process for the detection, turns a ‘blind eye’ to the knowledge of CSI - thereby evading the pilot-overhead involved in channel estimation. This approach may make the design more spectrally efficient.

[0128] In orderto attain near-capacity performance, channel coding is commonly employed, such that soft information, in the form of log-likelihood ratios, are exchanged between the different receiver blocks. A machine learning approach may also be used, such that the MIMO soft decoding is performed without explicit CSI knowledge. The receiver relying on the learning strategy employed during the training process for the detection, turns a ‘blind eye’ to the knowledge of CSI - thereby evading the pilot-overhead involved in channel estimation. This approach may make the design more spectrally efficient.

[0129] Achieving high performance, high throughput multi-user communications in the mmWave spectrum calls for advanced transceiver design including the concept of multi-functional (MF)-MIMO combining the benefits of multiplexing, diversity and beamforming, which may be combined with the concept of multidimensional index modulation for further improvements. However, the conventional coherent designs employ maximum likelihood (ML) detection, relying on the knowledge of the channel information at the transmitter and/or receiver. However, estimating the channel information calls for sending pilot data that reduce the bandwidth efficiency in addition to the fact that channel estimation techniques are employed at the receiver, which adds to the receiver complexity. Additionally, the maximum likelihood has excessive complexity, which prevents ML from practical implementation.

[0130] It would be desirable to provide lower-complexity MIMO reception schemes, such as detection and soft decoding, eliminating channel estimation and therefore reducing or avoiding the overhead associated with channel estimation and channel state information (CSI) feedback, hence improving the spectral efficiency.

System Model

Transmission Model

[0131] FIG. 2 is a schematic illustration showing an example MS-STSK encoder according to some embodiments. Consider the system model shown in FIG. 2, such that the transmitter is equipped with N_t antenna arrays (AA) of K antenna elements (AE) each. In FIG. 2, the transmitter employs multi-set space- time shift keying (MS-STSK) scheme, where the information is conveyed by both STSK symbol and antenna combination (AC) information. In FIG. 2, the MS-STSK scheme uses M RF chains 210, where the AC information selects M AA out of N_t total AA. More explicitly, the MS-STSK codeword includes two parts, where the first sequence of log₂(M_cM_Q ), where M_c is the constellation order and MQ is the number of dispersion matrices. An input stream / packet of bits 202 are mapped using STSK encoder 204 of FIG. 2, while the ensuing bits are used to convey the information of the AC to the antenna selection unit

206 of FIG. 2. For some embodiments, the ST Mapper 208 maps STSK encoded bits and antenna selection information to a set of M RF chains 210. During the MS-STSK transmission, only M antenna arrays 214 are activated at any time, while the other antennas from the set of N_t antennas 216 remain silent. An RF switch 212 may be used to activate the selected set of M antenna arrays 214. Each set of M is disjoint with the other sets. In an alternate embodiment, non-disjoint sets are permitted, leading to a maximum of {N_t choose M) sets.

[0132] FIG. 3 is a schematic illustration showing an example beamformed MS-STSK symbol transmitted in the direction of a desired user according to some embodiments. During transmission, an MS-STSK symbol is formed when the STSK symbol is inputted to the ST mapper 208 of FIG. 2, where a specific AC is selected depending upon the input bit-sequence. A part of the input bit-sequence indicates the AC to be selected for transmission. For some embodiments, the MS-STSK symbol 302 may be generated with the input stream / packet of bits 202, STSK encoder 204, antenna selection unit 206, ST Mapper 208, and M RF chains 210 of FIG. 2. An RF switch 304 may be used to activate a set of M antenna arrays 306. The MS-STSK symbol 302 is steered over a mmWave channel 310 using an RF analog beamformed (BF) matrix FRF using the preferred beam 308 for the desired user 314, as shown in FIG. 3. At the receiver, the received signal is used during the detection process, where conventionally detection of the MS-STSK symbol is performed with ML relying on the CSI at the receiver, which means a full search over the N_r antenna arrays 312 for the AC index and the dispersion matrix (DM) index as well as the modulation used.

[0133] The resultant codeword X transmitted by an MS-STSK encoder may be given by Eq. 1 :

X = A_qx_l Eq. 1 where x_l is the M-QAM/PSK symbol, and A_q is the q-th dispersion matrix of size M x T from the set A = {A₁ ...,A_q, ...,A_MQ }, where . The matrix A_q disperses the symbol x_l over M AA in T

timeslots. For some embodiments, for an input bit-sequence of ‘0110’, the first two bits, ‘1 O’, may be mapped to a 4-QAM symbol, while the remaining two bits, 01, may be mapped to one of four dispersion matrices from the set A with cardinality 4 {A = {A₁ A₂, A₃, A₄}). For some embodiments, the first three bits may be mapped to an 8-QAM symbol, while the last bit may be used for the selection of one of two dispersion matrices depending on the design requirements. For some embodiments, by using MS-STSK transmission, a total of log₂(N_cM_QM_c ) bits may be used for conveying the information.

[0134] Consider an example, where N_c = 2, M_Q = 4, and M_c = 4. In other words, the transmitter is equipped with 2 ACs and transmitting 4-QAM symbols with the aid of 4 dispersion matrices. Thus, given the above MS-STSK parameters, a total of log₂( 2 x 4 x 4) = 5 bits can be transmitted. As a design example, let us assume having the input bit sequence of ‘0110T. Then, the mapping of the bit sequence to MS-STSK is carried out as follows. By considering the least significant bit first, the bit T is mapped to the transmit antenna index, while the ensuing bits ‘10’ are mapped to one of the four dispersion matrices. And finally, the bits ΌT are mapped to a 4-QAM symbol.

[0135] Thereafter, the MS-STSK codeword of Eq. 1 is transmitted over mmWave channel by employing beamforming. The block-based received signal Y at the receiver after analog RF combining is given by:

Y = W_RFHF_rfX + V Eq. 2 where V is the Gaussian noise, and F_RF and W_RF are the analog RF beamforming and combining matrices of sizes KN_t x N_t and N_r x KN_r, respectively. They are expressed as:

F_RF = [ 0 - F_RFq ... 0] Î c^KNtx^Nt Eq. 3

F_RFq diag( ¹ _RF ² _RF F_rf) Eq. 4 where Fⁱ _RF is the BF vector of the i-th AA of size K x 1. Similarly: W_RF = diag(W_RF1 W_RFi - W_RFNr)^T Î C^Nr^xKNr Eq. 5 where W_R ^l _F is the BF vector of i -th AA of size K x 1. Furthermore, H is the channel matrix of size KN_r x KN_t expressed as:

H = [H₁ H₂ - H_Nc] Eq. 6 where H_t is the sub-channel matrix of size N_r K x MK, which is expressed as:

while H_J ^m is the mmWave channel matrix of size K x K spanning between the j-th AA at the receiver and the m-th AA at the transmitter. A statistical channel model having N_c clusters with N_ray each, may be given by

[0136] For a uniform linear AAs, the response vectors a_r and a_t are given by

where fr and ft are the angles of arrival and departure, respectively.

At the receiver, Eg. (2) is vectorized during the detection stage, which is given by y = HXI_cK_q,l + V Eq. 11 where each constituent vectorized matrix is expressed as

[0137] Furthermore, in the conventional MS-STSK aided transceiver design, soft-decision detection is carried out by outputting the log-likelihood ratio (LLR) of the MS-STSK demodulator. The LLR of a bit is defined as the ratio of the probabilities associated with the logical bit Ό’ and logical bit ‘T, which is formulated as:

where p(b = 1) and p(b = 0) are the probabilities associated with the logical bit '1 ' and logical bit '0', respectively. The sign of L(b) indicates the logical bit '1 ' or '0', while the magnitude indicates the confidence in that specific bit. [0138] The probability of receiving the signal y given that the symbol K_{q l} is transmitted from the c-th AC is given by:

On the other hand, the received signal y conveys the bit sequence B = [b₁ ··· b_Nb] , where N_b = log₂(N_cM_QM_c ). The LLR of the bit b_i is given by:

where

and

[0139] Eq. 19 relies on the knowledge of CSI at the receiver. This requires pilots for channel estimation, hence reducing the effective data rate [20]. The effective Discrete-input Continuous-output Memoryless Channel (DCMC) capacity accounting for the pilot density f_p, which is the ratio of the number of pilots to the number of data symbols, is given by [21], [22]:

while

where

For N_d number of symbols and N_p number of pilots in a frame, f_p is expressed as

Furthermore, when the minimum mean squared error (MMSE) channel estimate is considered, the channel estimate error variance (s² _h ) for a total transmission signal power p_t is given by [23]

[0140] FIG. 4 is a schematic illustration showing an example beamformed MS-STSK symbol coupled with the beam index transmitted in the direction of a desired user according to some embodiments. In contrast to FIG. 3, FIG. 4 shows the MS-STSK symbol 402 coupled with the beam index before the final transmission. In the model shown in FIG. 4, information is conveyed by the index of the beam in addition to the information conveyed by MS-STSK symbol 402. More explicitly, if the channel 412 supports a plurality of beams (say N_b beams), the transmitter selects a specific beam for transmission depending upon the input bit-sequence, instead of transmitting all the beams at once. Thus, MS-STSK transmission with beam index modulation (BIM) is capable of achieving an additional bit rate of log₂(N_b ) bits per channel over its counterpart, which lacks BIM, at the cost of SNR. Similarly, in this case, maximum likelihood detection is employed, which relies heavily on the accurate channel information knowledge at the receiver. For some embodiments, each antenna in an antenna array 406 may be capable of generating N beams 408, 410. Similar to FIG. 3, an RF switch 404 may be used to activate a set of M antenna arrays 406. As in FIG. 3, the receiver of a user 416 in FIG. 4 has N_r antenna arrays 414.

[0141] To illustrate further, consider another example such that the channel seen from each AA supports a total of 4 beams for transmission. In order to increase the spectral efficiency, the BIM is invoked by relying on the index of beam used for transmission. In other words, MS-STSK transmission can be carried out on one of four beams from each AA by allowing additional bits to convey the index of the beam. This philosophy holds only when there are more than one beam. If there is only one beam, then MS-STSK with only beamforming is used. Thus, the total number of bits per channel used when BIM is coupled with the MS- STSK example discussed above is ten (10), such that Bl = 1101 in binary, and MS-STSK = 110110 in binary.

[0142] Returning to FIG. 4, the channel seen from each AA at the transmitter supports N_b beams, and only one of the N_b beams is selected for transmission depending on the bit-sequence. Thus, there are N_b possible combinations of beamforming for MS-STSK symbol transmission totaling in log₂(N_bN_cM_cM_Q ) information bits in contrast to the log₂(N_cM_cM_Q ) bits of MS-STSK’s transmission dispensing with beam index mode. And the block-based received signal Y_BI in this scenario can be expressed as

[0143] where H_BI is the statistical channel model of (8) in the n-th beam, while the sizes of H_BI, W_nRF F_nRF and X are the same, as described above. The only difference between Eqs. 2 and 28 lies in the manner of leveraging the beams made available by the channel. Similar to Eq. 12, Eq. 28 can be vectorized for the n-th beam. At the receiver, the vectorized received signal y_BI is used during the detection process. In this setting, the detection of the estimates of Eq. 2 is obtained by employing ML relying on the CSI

of the n-th beam at the receiver and it is expressed as

[0144] Eq. 29 is heavily reliant on accurate CSI for the successful detection of symbol indices, thereby imposing both the pilot overhead for channel estimation and the complexity in the system design. Furthermore, Eq. 29 produces an error floor when the CSI estimate error variance is set to 0.25. Additionally, the DCMC capacity of the MS-STSK with BIM is expressed as

where

[0145] By taking into account the pilot percentage f_p, the effective DCMC capacity is:

[0146] In alternative embodiments, modulation schemes other than MS-STSK are used. One such alternative includes the use of orthogonal frequency division multiple access (OFDMA). In such an embodiment, a receive vector y is formed representing the real and imaginary constituent components once the received symbols have been downconverted and transformed to the frequency domain via fast fourier transform (FFT). The receive vector y may incorporate aspects of the above-referenced signal processing such as transmit beamforming and analog combining at the receiver, as is known in the art.

Deep Learning Model

[0147] FIG. 5 is a schematic illustration showing an example neural network model according to some embodiments. Artificial Neural Networks (ANN) is a computational model inspired by the structural and functional aspects of biological neural networks. The model of a typical neural network is shown in FIG. 5, where a neural network like that of FIG. 5 has multiple layers. The first and the last layers are the input layer 502 and the output layer 526, and the layers between them are referred to as hidden layers 512. To elaborate more, the input vector x_l 504 is inputted to the input layer 502, which is connected to the hidden layer 1 of FIG. 5. The output from each neuron of the hidden layer is governed by an activating function f{). An activating function limits the amplitude of the output of a neuron. Typically, the activating function is a function with a weight vector and a bias.

[0148] The use of an ANN includes two stages: a training phase and a testing phase. In the training phase, known input and target output samples are used to compute weight matrices and bias vectors. The weights and biases are designed in such a way that they minimize the error between the target output and the predicted output. In the testing phase, the pre-designed weights and biases are applied to new input data (outside the training samples) to predict the output.

[0149] The training weights and biases are determined for the neural network to be used with the detector. In an exemplary design, the number of hidden layers is set to 2, and the number of neurons is adjusted in such a way that the neural network reproduces faithfully the output during the training stage without using an excessive number of neurons. In a MIMO symbol transmission, the received symbol vector y serves as the input vector x_i 504 to the neural network, while the detected dispersion matrix index, the antenna index, and the complex-valued symbol drawn from M-QAM constellation constitute the output vector. Similarly, when an MS-STSK encoder is amalgamated with a beam-index, the received real and imaginary components of the symbol on multiple Rx antennas serves as the input to the ANN. Note that the input vector size may be scaled by the oversampling rate of the receiver. In this scenario, we have an additional element, which is the beam index, at the output of the ANN.

[0150] Having identified the input and the output vectors of the neural network, training of the network may be conducted with known input and target output samples. Before the training process is performed, the weight matrices and the biases vectors of all the layers are set to random values. Furthermore, a hyperbolic tangent function, known as a type of sigmoid, is used as the activating function because of its smoothness and asymptotic properties. However, for some embodiments, other activation functions may be used. The activating function is applied at each neuron of the network whose output is sent as the input to the next layer of neurons. Each activating function 506, 514, 520 maps its respective input vector x_i u_i or v_i of the i- th training sample using the weight matrix W 508, 516, 522 and bias vector b 510, 518, 524 of that layer. This mapping is used as the input vector to its succeeding layer and so forth. After the final mapping, which is at the output layer 526 with output vectors y₁ 528 and y₂ 530, the error is computed between the known output and the predicted output.

[0151] The weight matrices and bias vectors are designed in such a way that they minimize the loss function. The channel is assumed to be correlated over time and evolves, e.g., according to Jakes’ model, where the channel’s correlation coefficient in time is defined as shown in Eq. 33: z = J₀(2pf_dT) Eq. 33 where f_d is the maximum Doppler frequency and t is the sample time. The Doppler spread (f_dt) plays a role in deciding how often the training of the network parameters, e.g., weights and biases, are used in order to accurately estimate the indices.

[0152] After designing the neural network parameters, the testing phase is ensued, where the vector y from the receive AA is inputted to the input of the neural network. Here the parameters computed during the training phase are applied on the input vector to estimate the indices at the output of the network. Note that the input of the neural network takes only real values; therefore, the received vector y (where y= Real(y)-H lmag(y)) is split into two real parts from the constituent parts, Real(y), and, lmag(y), before feeding the two real parts or components, Real(y), lmag(y), to the neural network.

[0153] The output vector may take multiple forms and generally offer a trade-off between complexity and performance. Consider for example just one index component of the output which has, e.g., K = 16 possible values. This index could be encoded as a single scalar withl 6 different possible values, e.g., the example of Eq. 34:

[-7.5, - 6.5, ... , 7.5] * c Eq. 34 where c is some scale factor. In this case, the ANN is configured to solve a regression problem, and mean squared error (MSE) is one example loss function that could be used during training. The selected index is just the index of the value that is closest to the ANN output. Another alternative is to use a binary representation of the index requiring ceil(log₂(K)) outputs (generally carrying 1 bit each). In this case, the sign of each output is taken to obtain the bit value of each. Another alternative is to use so-called “one-hot” encoding. In this case, there is one output per possible value, K = 16 in this example. One-hot encoding may be coupled with the softmax() function, which forces the vector output to be a probability distribution (bounded to the interval [0,1] and summing to 1.0). This type of encoding has shown to be beneficial for performance. In this configuration, the ANN is a classifier and trained with a cross-entropy loss function. The same may be applied to each component of the output individually or collectively. Two indices with K1 = K2 = 4 possible values may be encoded and trained as a single K = 16 scalar, a 4-bit binary vector, or a 16-class classifier. Alternatively, the ANN may have two scalar outputs, each with 4 possible values.

Deep Learning Aided Semi-Blind Detection

[0154] FIG. 6A is a schematic illustration showing a first example of ANN assisted blind detection at the receiver according to some embodiments. FIG. 6B is a schematic illustration showing a second example of

ANN assisted blind detection at the receiver according to some embodiments. Because this design may estimate the output without channel state information (CSI), the pilot overhead may be reduced markedly. [0155] As discussed in the previous section, the training weights and biases will be determined for a neural network. For a learning aided blind-detection scheme, the number of hidden layers may be set to 2, while the number of neurons is adjusted in such a way that the scheme reproduces faithfully the target output during the training stage. For a MS-STSK symbol transmission, the vectorized matrix y 602 serves as the input to the neural network 604; while the complex-valued symbol drawn from M-QAM constellation (I) 606, the detected dispersion matrix index (q) 608, and the antenna index (c) 610 constitute the output vector, as shown in FIG. 6A. Similarly, if an MS-STSK encoder is amalgamated with beam index modulation, the vectorized matrix y_BI 652 is the input to the ANN 654. In this scenario, the complex-valued symbol drawn from M-QAM constellation (I) 656, the detected dispersion matrix index (q) 658, the antenna index (c) 660, and an additional element, which is the beam index (n) 662, are produced at the output of the ANN, as shown in FIG. 6B.

[0156] FIG. 7 is a functional block diagram illustrating an example of conventional maximum-likelihood detection for MS-STSK transmissions. FIG. 7 shows the block diagram of a typical receiver that uses Maximum Likelihood (ML) detection. On the transmitter side, a bit sequence 702 is transmitted with an MS- STSK transmitter 704 by an antenna array 706 over a mmWave channel 710. For this configuration of an ML system, pilots 708 are used to estimate the channel response matrix. In this design, the receiver first combines the signal spread over an antenna array 712 of N_r antennas in the RF stage and then performs down-conversion 714 for further digital processing in the baseband. Given the necessity of having the CSI, channel estimation 716 is carried out with the aid of pilots prior to the detection. ML detection 718 estimates the MS-STSK symbol index estimates 720 only after the CSI estimate is obtained.

[0157] FIG. 8 is a functional block diagram illustrating an example of learning assisted detection for MS- STSK transmissions according to some embodiments. FIG. 8 shows the process flow for a deep-learning aided transmitter design that does not send any pilots dedicated for channel estimation. On the transmitter side, a bit sequence 802 is transmitted with an MS-STSK transmitter 804 by an antenna array 806 over a mmWave channel 808. At the receiver, deep-learning aided detection may be performed without channel estimation. The received signal from the antenna array 810 is combined and down-converted 812. The down- conversion output is an input to an ANN 814, where the NN parameters learned during the training phase are applied for estimating the MS-STSK indices 816.

[0158] FIG. 9 is a message sequencing diagram illustrating an example process of deep learning detection for MS-STSK transmissions according to some embodiments. FIG. 10 is a message sequencing diagram illustrating an example process of deep learning detection for MS-STSK transmissions with beam index modulation according to some embodiments. FIGs. 9 and 10 show the signaling between the gNB 902, 1002 and the UE 904, 1004 for the deep learning aided detection with or without BIM, where after establishing the beam alignment, the setup of the BIM is performed and then the deep learning training phase is done followed by the testing phase.

[0159] For the example process shown in FIG. 9, the gNB 902 establishes 906 beam alignment with the UE 904. The example process continues with the gNB 902 sending 908 training data to the UE 904. The UE 904 trains 910 the neural network for detection. Upon completion, the UE 904 acknowledges 912 training is finished to the gNB 902. The example process continues with the gNB 902 sending 914 data via real data transmission to the UE. The UE 904 checks 916 to see if the key performance indicator (KPI) is greater than a pre-defined threshold. For some embodiments, a KPI may be, e.g., the Received Signal Strength (RSS), signal to interference plus noise ratio, and/or a reliability metric, such as bit error rate or packet error rate. If no, the UE may request 918 retraining and after which, the gNB 902 may send 908 training data to the UE 904. If yes, the UE 904 may continue receiving data 920.

[0160] For the example process shown in FIG. 10, the gNB 1002 sends 1006 beam alignment data to the UE 1004. The gNB 1002 communicates with the UE 1004 to establish 1008 valid beams and set up a BIM lookup tale. The example process continues with the gNB 1002 sending 1010 training data to the UE 1004. The UE 1004 trains 1012 the neural network for detection. Upon completion, the UE 1004 acknowledges 1014 training is finished to the gNB 1002. The example process continues with the gNB 1002 sending 1016 data via real data transmission to the UE. The UE 1004 checks 1018 to see if the KPI is greater than a predefined threshold. If no, the UE may request 1020 retraining and after which, the gNB 1002 may send 1010 training data to the UE 1004. If yes, the UE 1004 may continue receiving data 1022.

[0161] In terms of search space complexity, in the conventional scenario, the ML detection operates to estimate the index of the dispersion matrix, of the M-QAM symbol, and of the AC. Thus, the run-time complexity relying on ML detection would be as shown in Eq. 35 using the on the order of operator, O():

O(M_CM_QN_c ) Eq. 35 where there are M_Q dispersion matrices, M_c complex-valued symbols, and N_c antenna combinations. Furthermore, the ML-aided detection uses CSI, which imposes additional complexity during the channel estimation stage (shown in FIG. 7), while also significantly affecting the data rate because of the pilot consumption of the spectral resources. In contrast to the ML detection, learning-assisted detection shown in FIG. 8 improves the data rate by circumventing the necessity of having CSI at the receiver. A neural network, once trained, turns a “blind eye” to the CSI. This approach makes a learning-assisted design spectrally efficient. [0162] The complexity of the learning-assisted design depends on the number of neurons in each hidden layer. The complexity of a typical NN arises in two stages: forward propagation and backward propagation. To elaborate a little further, suppose that there are n neurons in a hidden layer. Assume that the input and output vectors are of sizes n_i and n_o, respectively. For each layer, the activating function is computed using the network parameters of the respective layer. The pre-determined weight matrix and bias vector values are substituted in the activating function having input vector ^ to compute the intermediary output n_o, which serves as the input to the next layer. Contrasting the learning-assisted design with an ML receiver’s search complexity by considering each search operation as a node of FIG. 6A or FIG. 6B, the complexity of the learning assisted design would be as shown in Eq. 36:

O(n_in_h1n_h2n_o ) Eq. 36 where n_i is the size of the input vector; n_h1 is the number of neurons, or size of the vector, at the hidden layer 1; n_h2 is the number of neurons, or size of the vector, at the hidden layer 2; and n₀ is the size of the output vector. In terms of number of computations, the total number of complex multiplications required in ML detection for the transmission scheme without BIM is shown in Eq. 37:

20{N_tN_rN_cM_QT ²) + 0(N_CM³ _Q) Eq. 37

For the transmission scheme with BIM, the total number of complex multiplications required in ML detection is shown in Eq. 38:

20(NN_tN_rN_cM_QT²) + 0(NN_cM³ _Q) Eq. 38

By contrast, for the NN with the aforementioned parameters, the number of real-number multiplications is as shown in Eq. 39:

where n_i is the size of the input vector; n_hi is the number of neurons, or size of the vector, at the hidden layer 1; n_h2 is the number of neurons, or size of the vector, at the hidden layer 2; and n₀ is the size of the output vector. Table 1 provides a comparison of the number of computations required for the MS-STSK with/without BIM for the example system parameters shown in Table 2. The computational complexity of conventional ML-based detection is similar to learning-aided detection. Furthermore, the conventional ML- based detection entails having channel estimation, which is eliminated in the learning-aided detection. Table 1. Complex Multiplications for Design Schemes

Table 2. System Parameters

[0163] FIG. 11 is a graph illustrating example bit error rates for MS-STSK transmissions according to some embodiments. FIG. 11 shows the bit error rate (BER) 1102 vs. signal-to-noise ratio (SNR) 1104 for BER of learning-assisted blind detection 1112, of ML-aided detection with perfect CSI 1106, and of ML-aided detection with imperfect CSI for the MS-STSK transmission dispensing with the BF index modulation 1108, 1110. In a setting with no associated BF index, the channel may be assumed to support only one beam, or all potential beams are utilized for the transmission. In this design, the neural network (NN) estimates the indices for the antenna, the dispersion matrix, and the symbol. The parameters used for the simulations of FIG. 11 are shown in Table 2.

[0164] As seen in FIG. 11 for the aforementioned NN parameters, for a BER of 10^-3, there is about a 6 dB gap in SNR between the learning-assisted detection 1112 and the ML-aided detection with perfect CSI 1106. Although the ML-aided detection with perfect CSI 1106 outperforms the learning-assisted detection 1112 by 6 dB SNR, the gain comes with an unrealistic assumption of having a perfect CSI. The SNRs shown in FIG. 11 are in the range of -30 to -10 dB because the MIMO system is characterized by diversity and beamforming gains that improve system performance significantly.

[0165] On the other hand, for the CSI with the error variance of 0.15, the SNR gap reduces to around

3 dB, whereas the ML-aided detection produces an error floor for the CSI error variance of 0.25. By contrast, the learning-assisted blind detection is able to estimate faithfully the indices of the MS-STSK transmission regardless of the nature of the CSI and circumvent pilot-assisted channel estimation.

[0166] Although FIG. 11 shows an SNR gap between the learning-assisted scheme 1112 and the ML- based detectionl 106, the capacity of both designs will be examined for the sake of fairness. The SNR gain observed for ML-assisted detection 1106 in FIG. 11 is contingent on the CSI estimation, whose accuracy increases in proportion to the pilots density. Flowever, increasing the pilot density reduces the overall capacity.

[0167] FIG. 12 is a graph illustrating example discrete-input continuous-output memoryless channel capacities for MS-STSK transmissions according to some embodiments. The parameters used for the simulations of FIG. 12 are shown in Table 2. FIG. 12 characterizes the Discrete-input Continuous-output Memoryless Channel (DCMC) capacity of MS-STSK 1202 vs. SNR 1204 for DCMC of the learning-assisted blind detection 1208 and for DCMC of the ML-aided detection for 0% pilot overhead 1206, 3% pilot overhead 1210, 5% pilot overhead 1212, and 10% pilot overhead 1214. As shown in FIG. 12, the capacity of the ML- based design is strictly limited by the percentage of the pilot overhead. More explicitly, for the simulation parameters summarized in Table 2, the capacity of the ML-aided detection for 3% pilot-overhead 1210 is limited to a maximum value of 4.85 bits per channel user (bpcu), while the capacity is 4.75 bpcu for 5% pilot overhead 1212. Furthermore, if the pilot overhead is increased to 10% (1214), the DCMC capacity of the ML- based detection is reduced to a maximum value of 4.5 bpcu. By contrast, the DCMC capacity of the deep learning-assisted detection 1208 reaches the maximum value of 4.99 bpcu, which is close to the DCMC capacity 1206 of 5 bpcu, because the overhead involved in recalibrating the weights is marginally less and employs detection of the MS-STSK symbol by turning a “blind eye” to the CSI. Additionally, there is a SNR gain of 0.5 dB and 1 dB between the learning-assisted detection and the ML-detection with 5% overhead for achieving the rate of 3 bpcu and 4 bpcu, respectively; while the SNR gain is around 1 dB and 3 dB between the learning-assisted detection and the ML detection with 10% overhead for achieving the rate of 3 bpcu and

4 bpcu, respectively.

[0168] FIG. 13 is a graph illustrating example bit error rates for MS-STSK transmissions with beam index modulation according to some embodiments. The parameters used for the simulations of FIG. 13 are shown in Table 2. FIG. 13 shows the BER 1302 vs. SNR 1304 for the BER of the learning-assisted blind detection 1312, of the ML-aided detection with perfect CSI 1306, and of the ML-aided detection with imperfect CSI 1308, 1310 when the MS-STSK transmission is coupled with beamforming (BF) index modulation (BIM). In this setting, we assumed that the channel supports two beams for each AA. There is an extra index, which is the beam index, for estimation. In this scenario, it is empirically observed during the training phase that the numbers of neurons may be set to 30 and 30 for both real and imaginary constituents of the NN. As seen in FIG. 13, adding an additional index for estimation increased the SNR gap between the learning-assisted blind detection 1312 and the ML-aided detection with perfect CSI 1306 to 8 dB. Again, the superior performance of the ML-based detection 1306 is because of the unrealistic assumption of having a perfect CSI. However, if the CSI estimate with the error variance of 0.15 is introduced, the ML-aided detection 1308 starts to produce an error floor from around - 10 dB, while the BER remains flat for the CSI error variance of 0.25 (1310). On the other hand, learning-assisted detection 1312, despite the absence of CSI, estimates both the MS-STSK indices and beam index with integrity.

[0169] FIG. 14 is a graph illustrating example discrete-input continuous-output memoryless channel capacities for MS-STSK transmissions with beam index modulation according to some embodiments. The parameters used for the simulations of FIG. 14 are shown in Table 2. FIG. 14 characterizes the DCMC capacity 1402 vs. SNR 1404 for the DCMC of the learning-assisted detection 1408 and for DCMC of the ML- aided detection for 0% pilot overhead 1406, 3% pilot overhead 1410, 5% pilot overhead 1412, and 10% pilot overhead 1414 when the MS-STSK transmission is amalgamated with BIM. In this simulation, the channel supports two beams, and BIM is used with these two beams such that only one beam is activated depending on the input bit-sequence. As shown in FIG. 14, the DCMC capacity of learning-assisted blind detection 1408 is better than ML-aided detection 1410, 1412, 1414. This is because of the overhead induced by the pilots for channel estimation to aid the ML detection process. This is especially more pronounced when the pilot overhead is set to 10% (1414) as seen in the figure, where the DCMC capacity is less than 5.5 bpcu while the learning-aided detection 1408 is 6 bpcu. The use of pilots to estimate the CSI partly consumes the physical resources, thereby reducing the capacity of the system forML detection 1410, 1412, 1414. However, this behavior is not observed in the learning-assisted scheme 1408, because the learning-assisted scheme 1408 conducts the symbol detection with overhead in retraining as low as 0.002%. Furthermore, we observe the SNR gain of 3 dB to achieve the rate of 5 bpcu between the learning-assisted detection 1408 and the ML-aided detection with 10% overhead 1414, while the SNR gain is around 2 dB to achieve the rate of 4 bpcu. Deep Learning Aided Semi-Blind Detection

[0170] FIG. 15 is a graph illustrating example bit error rates for different numbers of frames and Jake’s correlation coefficient values according to some embodiments. The BER of the learning-assisted design has been analyzed for different numbers of frames as the channel evolves in time according to the normalized Doppler frequency. FIG. 15 shows the BER 1502 vs. SNR 1504 for 6 combinations of the number of frames and the Doppler spread {f_dr). The graph shows BER traces for 10 frames with f_dr = 0.001 (1506), 50 frames with f_dt = 0.001 (1508), 100 frames with f_dt = 0.001 (1510), 10 frames with f_dt = 0.0005 (1512), 50 frames with f_dt = 0.0005 (1514), and 100 frames with f_dt = 0.0005 (1516). FIG. 15 shows that as number of frames increases from 10 to 100, the BER of the learning-assisted design degrades. This phenomenon is observed because the training weights designed during the first few frames become outdated in time after a certain number of frames; and hence retraining of the NN parameters is necessary. The number of frames before the NN weights become outdated may depend on the Doppler spread. For example, in FIG. 15, the number of frames before the BER degrades for the normalized Doppler spread ( f_dr ) of 0.0005 is larger than for the normalized Doppler spread value of 0.001. Therefore, for some embodiments, a recalibration mechanism may be introduced whereby the receiver (e.g., User Equipment) requests the transmitter (e.g., Base Station) to transmit pilots to recalibrate its weights depending on the observed BER.

[0171] FIG. 16A is a frame structure diagram illustrating an example maximum-likelihood (ML)-assisted detection according to some embodiments. FIG. 16B is a frame structure diagram illustrating an example learning-assisted detection according to some embodiments. FIGS. 16A and 16B show the schematic of the pilot transmission for both the learning-assisted design and ML-based detection, respectively. In ML-aided detection, pilots (shown as “P” in FIG. 16A) 1602 and data 1604 are transmitted in every frame, whereas in learning-assisted detection (FIG. 16B), training data T_d 1652 is requested by the user only after Nt frames of data 1654, which may be contingent on the Doppler spread. For some embodiments, N_p in FIG. 16A may be larger or smaller than N_p in FIG. 16B. For some embodiments, the data field 1654 in FIG. 16B may have more bits than the sum of all the bits in the data fields 1604 in FIG. 16A. The pilots are transmitted for every frame with ML-based detection. By contrast, learning-assisted detection uses the training data for recalibrating the NN weights only after every Nf frames, as shown in FIG. 16B, while performing blind- detection in the rest of the frames. This may also be interpreted as online learning. Thus, the learning- assisted scheme performs in a semi-blind manner, because the learning-assisted scheme uses recalibration of its weights relying on the training data and performs blind-detection thereafter. It is notable that the amount of training data called for during recalibration decreases with the decrease in the Doppler spread. The overhead involved for retraining the NN network of an embodiment is shown in Eq. 40:

where Nf is the number of frames, N_d is the number of data streams, and N_P is the number of pilots; while the pilot overhead involved in the channel estimation for ML-aided detection is given by Eq. 41:

Deep Learning Aided Soft Detection

[0172] FIG. 17 is a schematic illustration showing example inputs and outputs of the ANN for soft detection according to some embodiments. This section applies the deep learning aided semi-blind design to soft detection. Note that the rationale for choosing learning over conventional soft detection is that by employing the former the requirement of pilot-assisted channel estimation may be eliminated. Unlike maximum likelihood (ML), in which an exhaustive search is performed over all possible combinations of the bit being either logical 0 or 1, the learning-aided design provides the soft log likelihood ratios (LLRs) by using ANN weights designed during the training phase. In this design, the input training samples of the ANN 1704 are the received signal vectors y 1702, whereas the target output training samples are the LLRs 1706, as shown in FIG. 17. The ANN 1704 is trained to infer the functional mapping between the input and output samples. Flowever, because the received signal vectors are affected with the noise, which is random, the ANN 1704 may not accurately infer the function. Therefore, the choice of the SNR during the training, which is obtained empirically, may be a significant factor. For a given SNR, the ANN 1704 predicts the LLR value by using the activation function for each input training sample.

[0173] If an exhaustive search is carried out over all the legitimate combinations of the bit being either logical Ό’ or ‘T, the learning-aided design provides the soft-LLRs by employing the ANN weights designed during the training phase. In this design, the input training samples of the ANN are the received signal vectors y, while the output labels are the LLRs, as shown in FIG. 17. Then, the ANN is trained to infer the functional mapping between the input and output samples. Flowever, since the received signal vectors are affected by the noise, the ANN may fail to accurately infer the function. Therefore, the choice of the SNR during the training is not insignificant, and in some embodiments may be obtained empirically by varying the SNR.

[0174] For a given SNR, the ANN predicts the LLR value by employing the AF f{v) in each layer of the network, where the input of each AF is the output of the preceding layer. For some embodiments, a sigmoid function is used as the AF to benefit from its smoothness. The sigmoid function may be formulated as:

[0175] For some embodiments, a sigmoid function may be expressed as:

where /(x, a, v ) is a mapping of x, a is the slope parameter, and c affects the function position. [0176] For example, the input of the first neuron in the second layer may be expressed as:

[0177] The AF of Eq. 42 is applied to v₁ to obtain a mapping, which serves as one of the inputs to the next layer. In this way, the output of the AF of each layer is then fed to its subsequent layer as shown in FIG. 17. This process is carried on until the output layer is reached, where the final predicted values are obtained.

[0178] Note that initially the weights of each layer are assigned to random values obeying the distribution JV( 0,1) . These weights are then updated for ensuring that they minimize the error between the predicted LLR and the actual LLR. Mathematically, it is formulated by a loss function (LF) given by:

where S is the cardinality of the training set, and L_i are the predicted and the known LLR values,

respectively, of the i -th training sample, while p₁, p₂, p₃ are the regularization factors used to avoid over- fitting, according to C. M. Bishop, Pattern Recognition and Machine Learning, INFORMATION SCIENCE AND STATISTICS (2006).

[0179] To minimize the loss function of Eq. 45, the gradient of the loss with respect to the weights is computed and used for updating the weight values in a gradient descent procedure known as back- propagation. These weights, which are learned during the training phase are then stored in memory and are invoked during the testing phase. In other words, the ANN predicts the LLR value from the received signal vector y by employing the pre-determined weights. Thus, for some embodiments, knowledge of the channel is not used to obtain the LLR values. During the training of ANN weights, the AoAs and AoDs of the channel matrix of Eq. 8 are assumed to be time-unvarying, while the small-scale fading coefficient is assumed to evolve in time according to Jakes’ model. The soft-LLR values predicted from the MS-STSK’s ANN demodulator are then passed to the turbo channel decoder.

[0180] FIG. 18 is a functional block diagram illustrating an example ML assisted detection for MS-STSK transmissions. FIG. 18 shows the block diagram of a conventional MS-STSK soft-demodulator. On the transmitter side, a bit sequence 1802 is encoded by an FEC encoder 1804. The encoded bits are transmitted with an MS-STSK transmitter 1806 by an antenna array 1808 over a mmWave channel 1810. For this configuration of an ML system, pilots 1812 are used to estimate the channel response matrix. The analog signal received over an antenna array 1814 is RF combined and down-converted 1816 to baseband. The receiver estimates the channel matrix H 1818 with the aid of pilots. Upon estimating the channel, the receiver uses a soft-demodulator 1820 to obtain the LLR values, which are passed to the ensuing FEC channel decoder 1822. The channel decoder 1822 iteratively improves the confidence of the LLRs and outputs the uncoded bits 1824. In this design, the receiver operates to estimate the channel for every frame of data. This design entails the transmission of pilots in every data frame for acquiring the channel estimate. Thus, this design imposes pilot-overhead and reduces the overall system throughput. FIG. 18 shows a turbo, forward error correction (FEC) encoder 1804 / decoder 1822 as an example channel encoder/decoder combination, but any channel encoder/decoder combination may be used with a learning-assisted demodulator.

[0181] FIG. 19 is a functional block diagram illustrating an example deep learning assisted detection for MS-STSK transmissions according to some embodiments. FIG. 19 shows the deep-learning-assisted soft- demodulator. On the transmitter side, a bit sequence 1902 is encoded by an FEC encoder 1904. The encoded bits are transmitted with an MS-STSK transmitter 1906 by an antenna array 1908 over a mmWave channel 1910. Like the conventional design, the signal received over an antenna array 1912 is analog RF combined and down-converted 1914. However, in contrast with the conventional design, the learning-aided soft-demodulator does not require the knowledge of the CSI. The down-converted signal vector is input to the ANN 1916, which uses pre-determined weights to obtain the LLRs 1918 without any requirement of CSI. The learning-assisted design does not use a channel estimation stage, which persists in the conventional design. The LLRs 1918 from the ANN 1916 are passed to the channel decoder 1920, and the channel decoder 1920 outputs the uncoded bits 1922. FIG. 19 shows a turbo, forward error correction (FEC) encoder 1904 / decoder 1920 as an example channel encoder/decoder combination, but any channel encoder/decoder combination may be used with a learning-assisted demodulator.

[0182] To illustrate the complexity quantitatively, assume that input and output vectors of the ANN shown in FIG. 15 are of sizes n_i and n_o, respectively. Also, suppose that the number of neurons in each hidden layer as n_h. We know that the AF is computed for each neuron of each layer by employing the corresponding layer’s weights and biases. Then the total number of computations used for the ANN of FIG. 15 is shown in Eq. 46:

The number of computations used for the conventional soft-demodulator is shown in Eq. 47:

0(N_tN_rN_cM_QT ²) + 0(N_CM³ _Q) Eq. 47 Table 3 provides a comparison of the number of computations used for the MS-STSK for the system parameters shown in Table 4.

Table 3. Complex Multiplications for Soft Decoding Design Schemes

Table 4. System Parameters

[0183] FIG. 20 is a graph illustrating example bit error rates for conventional and learning aided soft demodulation according to some embodiments. The parameters used for generating these simulations are shown in Table 4. FIG. 20 characterizes the bit error rate (BER) 2002 vs. SNR 2004 for the BER of the learning-aided soft-demodulator 2006 and of the conventional MS-STSK soft-demodulator 2008. As shown in FIG. 20, despite its unawareness to the CSI knowledge, the learning-aided soft-demodulator 2006 performs closely to the conventional soft-demodulation 2008. More particularly, at BER < 10^-4, the SNR gap between the two is 1.5 dB. Example embodiments of a learning-aided soft-demodulator circumvent the pilot overhead, channel estimation, and the increased search complexity of the conventional MS-STSK demodulator.

[0184] FIG. 21 is a graph illustrating an example discrete-input continuous-output memoryless channel capacities for conventional and learning aided soft demodulation for 3%, 5%, and 10% pilot overheads according to some embodiments. The parameters used for generating these simulations are shown in Table 4. FIG. 21 shows the DCMC of MS-STSK 2102 vs. SNR 2104 for the DCMC of the learning-assisted detection 2108 and for DCMC of the conventional soft demodulator without pilot overhead 2106, 3% pilot overhead 2110, 5% pilot overhead 2112, and 10% pilot overhead 2114. While learning-assisted soft demodulation performs within 1.5 dB to achieve a BER of 10^_4 as shown in FIG. 20, learning-assisted soft demodulation 2108 provides a higher Discrete-input Continuous-output Memoryless Channel (DCMC) capacity than conventional soft demodulation 2110, 2112, 2114, as shown in FIG. 21. This is because of the pilot overhead used for channel estimation to carry out the soft detection. Furthermore, the pilot overhead for conventional design may span from 3% to 10% of the data rate, depending on the Doppler spread. By contrast, the learning-assisted soft-demodulator does not entail channel estimation, thereby, precluding the pilot overhead while providing higher data rates. As shown in FIG. 21, the learning assisted design provides an SNR gain of 3 dB when compared with the conventional design with 10% overhead at a rate of 4 bps/Hz, while the SNR gain is around 1.5 dB and 0.8 dB when having 5% and 3% pilot overhead, respectively.

[0185] In conclusion, in contrast to the conventional MS-STSK soft-demodulator, which entails the knowledge of having CSI at the receiver, the example learning-assisted design does not use channel estimation and improves the data rate, as the latter does not impose pilot overhead. Furthermore, the learning-aided soft-demodulation avoids the exhaustive search complexity for evaluating the soft values at the output of the MS-STSK demodulator. Despite both its incognizance about the CSI and the low complexity, the learning-assisted design performs closely to the conventional design assuming the perfect CSI for BER < 10 ⁴, whereas in the event of imperfections in the CSI at the receiver for conventional soft-demodulation, the learning-aided soft-demodulator outperforms the latter.

Fading Memory Experience Replay

[0186] In order to reduce or eliminate the need to send pilots for re-training the NN after the channel changes, an experience replay buffer may be created from received packets.

Creation of Replay Buffer

[0187] Input samples are saved in sample buffer. The sample buffer is large enough to hold samples until the corresponding packet is checked, e.g., by CRC. After a packet is checked, the corresponding samples are either discarded (if CRC fails) or placed in the replay buffer. If placed in the replay buffer, the following may also be placed in the replay buffer: 1) a time stamp indicating when the samples were collected or how long the samples will stay in the replay buffer; 2) the packet is re-encoded to reproduce the transmitter output and corresponding target output of the NN. The target output is also saved in the replay buffer as a label of the input samples and will be used for training. An estimate of the channel coherence time (and possibly other metrics such as recent BER or PER) is used to determine how long each replay buffer entry will remain in the replay buffer. Entries older than a threshold are removed. In the case that replay buffer will overflow, the oldest entry is also removed to make room for new data. If the amount of data in the buffer falls below a second threshold, the lifetime of the reaming data is extended. Note, the second threshold may be zero.

Training with Replay Buffer

[0188] As long as there is sufficient data in the replay buffer, the NN may continuously learn online and track the channel conditions. In one embodiment, a parallel NN is instantiated and the weights of the 1^st NN are copied to the 2^nd NN after initial training. The 2^nd NN trains using the replay buffer while the 1^st NN receives user data. When triggered, the weights of the 2^nd NN are copied into the 1^st NN, cloning it at that instant. The triggers may be 1) periodic, possibly based on the estimated coherence time; 2) based on a measurement of how much the weights of the 2^nd NN have changed since the last copy; 3) an estimate that the 1^st NN performance has degraded; 4) received data is periodically send to both the 1^st and 2^nd NN and the CRC for both is compared. If the CRC for the 2^nd NN passes but fails for the 1^st NN a trigger is created after some number >0 of such events. The triggers may be delayed until the 1^st network is idle, e.g., between receiving packets.

[0189] In another embodiment, only the 1^st NN is instantiated, and the NN uses otherwise idle period to train using the replay buffer. In one embodiment, the data in the replay buffer is accessed sequentially and simply wraps around at the end of the buffer or otherwise accesses data with uniform probability. In an alternative, newer data has a higher probability of being accessed for training. Regardless of the probability distribution of data access, each epoch may be randomized.

[0190] FIG. 22 is a flowchart illustrating an example process for obtaining uncoded bits using an ANN according to some embodiments. For some embodiments, an example process 2200 may include receiving 2202 a signal and performing analog RF combining. For some embodiments, the example process 2200 may further include down-converting 2204 the combined received signal. For some embodiments, the example process 2200 may further include providing 2206 the down-converted combined signal to an artificial neural network (ANN). For some embodiments, the example process 2200 may further include demodulating 2208 the received signal using predetermined weights in absence of channel state information to obtain uncoded bits.

[0191] For some embodiments, an apparatus may include: a processor; and a non-transitory computer- readable medium storing instructions operative, when executed by the processor, to: receive a signal and performing analog RF combining; down-convert the combined received signal; provide the down-converted combined signal to an artificial neural network (ANN); and demodulate the received signal using predetermined weights in absence of channel state information to obtain uncoded bits. [0192] For some embodiments, an apparatus may include: an RF combiner; a downconverter; an artificial neural network (ANN) module; and a forward error correction (FEC) decoder. Some embodiments of an apparatus may further include a replay buffer.

[0193] FIG. 23 is a flowchart illustrating an example process for obtaining an ANN output vector according to some embodiments. For some embodiments, an example process 2300 may include using a vectorized matrix y comprising a received signal of received signal values as an input to an artificial neural network (ANN). For some embodiments, the example process may further include obtaining an ANN output vector comprising a detected dispersion matrix index, an antenna index, and a complex-valued symbol drawn from an M-QAM constellation.

[0194] An example method in accordance with some embodiments may include: using an artificial neural network (ANN) to decode a received signal, wherein a vectorized matrix y containing received signal values is used as an input to the ANN, and a detected dispersion matrix index, an antenna index, and a complexvalued symbol drawn from M-QAM constellation constitute an ANN output vector.

[0195] For some embodiments, an apparatus may include: a processor; and a non-transitory computer- readable medium storing instructions operative, when executed by the processor, to: use a vectorized matrix y comprising a received signal of received signal values as an input to an artificial neural network (ANN); and obtain an ANN output vector comprising a detected dispersion matrix index, an antenna index, and a complex-valued symbol drawn from an M-QAM constellation.

[0196] For some embodiments of the example method, a number of hidden layers in the ANN may be set to 2.

[0197] For some embodiments of the example method, a number of neurons in each level of the ANN may be adjusted to reproduce the target output during the training stage.

[0198] For some embodiments of the example method, the received signal may have been modulated at a transmitter using MS-STSK symbol transmission.

[0199] For some embodiments of the example method, the received signal may have been modulated at a transmitter with an MS-STSK encoder in combination with beam-index modulation, and a vectorized matrix yBi serves as the input to the ANN.

[0200] For some embodiments of the example method, the ANN output vector may further include a beam index. [0201] For some embodiments of the example method, the ANN may further include weights and biases determined during a training phase to minimize an error between a target output and a predicted output.

[0202] For some embodiments of the example method, the ANN may include activating functions of the form of a hyperbolic tangent function, or a type of sigmoid function.

[0203] For some embodiments of the example method, the received signal may be divided into real values by splitting the received vector y into real part Real(y) and imaginary part lmag(y) before providing it as an input to the ANN.

[0204] For some embodiments of the example method, training data for recalibrating the ANN weights may be sent after every Nf frames.

[0205] For some embodiments of the example method, blind-detection may be performed in frames received between training frames.

[0206] An additional example method in accordance with some embodiments may include: receiving a signal and performing analog RF combining; down-converting the combined received signal; providing the down-converted combined signal to an artificial neural network (ANN); using predetermined weights to obtain log likelihood ratios (LLRs) in the absence of channel state information; and processing the LLRs with a channel decoder to obtain uncoded bits.

[0207] Some embodiments of the example method or the additional example method may be performed in combination with a replay buffer.

[0208] For some embodiments of the additional example method, input samples may be saved in the replay buffer only if a cyclic redundancy check (CRC) passes.

[0209] Some embodiments of the additional example method may further include storing a time stamp indicating when the samples were collected or how long the samples will stay in the replay buffer.

[0210] For some embodiments of the additional example method, a packet from the replay buffer may be re-encoded to reproduce a transmitter output and corresponding target output of the ANN.

[0211] For some embodiments of the additional example method, any one of (i) an estimate of the channel coherence time, (ii) a recent bit error rate, or (iii) packet error rate may be used to determine how long each replay buffer entry will remain in the replay buffer.

[0212] A further example method in accordance with some embodiments may include: receiving a signal and performing analog RF combining to generate a combined received signal; down-converting the combined received signal; providing the down-converted combined signal to an artificial neural network (ANN); and using predetermined weights to demodulate the received signal in absence of channel state information to obtain data bits.

[0213] Some embodiments of the further example method may further include recalibrating periodically the predetermined weights using training update information transmitted from the transmitter to the receiver.

[0214] Some embodiments of the further example method may further include storing at least one of the data bits or the received, combined, and down-converted signal of received signal values as input samples to a replay buffer.

[0215] For some embodiments of the further example method, the input samples may be stored in the replay buffer only if a cyclic redundancy check (CRC) passes.

[0216] Some embodiments of the further example method may further include storing a time stamp indicating at least one of when the input samples were collected and how long the input samples will stay in the replay buffer.

[0217] For some embodiments of the further example method, a packet of uncoded bits from the replay buffer may be re-encoded to reproduce a transmitter output and corresponding target output of the ANN.

[0218] For some embodiments of the further example method, at least one of: (i) an estimate of a channel coherence time, (ii) a recent bit error rate, or (iii) a packet error rate, is used to determine how long at least one replay buffer entry will remain in the replay buffer.

[0219] For some embodiments of the further example method, obtaining uncoded bits may include: inputting to the ANN a vector y, representing a vectorized matrix, comprising the down-converted combined signal of received signal values; and obtaining an ANN output vector comprising a detected dispersion matrix index, an antenna index, and a complex-valued symbol drawn from an M-QAM constellation.

[0220] For some embodiments of the further example method, the ANN may be configured to have 2 hidden layers.

[0221] Some embodiments of the further example method may further include adjusting one or more neurons in at least one level of the ANN to reproduce a target output during a training stage.

[0222] For some embodiments of the further example method, the received signal may be modulated at a transmitter using MS-STSK symbol transmission.

[0223] For some embodiments of the further example method, the received signal may be modulated at a transmitter with an MS-STSK encoder in combination with beam index modulation, and the vectorized matrix y may be equal to a vectorized matrix y_BI corresponding to beam index modulation. [0224] For some embodiments of the further example method, the ANN output vector may further include a beam index.

[0225] For some embodiments of the further example method, the ANN may include weights and biases determined during a training stage to reduce an error between a target output and a predicted output.

[0226] For some embodiments of the further example method, the ANN may include activating functions of a type of hyperbolic tangent function, or a type of sigmoid function.

[0227] For some embodiments of the further example method, the received signal may be divided into real and imaginary values by splitting the received vector y into a real part Real(y) and an imaginary part lmag(y) before providing the real part Real(y) as the input to the ANN.

[0228] For some embodiments of the further example method, training data for recalibrating the ANN weights may be sent after every N_f frames.

[0229] For some embodiments of the further example method, blind-detection may be performed for frames received between training frames.

[0230] Some embodiments of the further example method may further include storing the received signal as input samples to a replay buffer.

[0231] A further example apparatus in accordance with some embodiments may include: a processor; and a non-transitory computer-readable medium storing instructions operative, when executed by the processor, to: receive a signal and perform analog RF combining; down-convert the combined received signal; provide the down-converted combined signal to an artificial neural network (ANN); and demodulate the received signal using predetermined weights in absence of channel state information to obtain uncoded bits.

[0232] A further additional example method in accordance with some embodiments may include: using a vector y, representing a vectorized matrix, comprising a received signal of received signal values as an input to an artificial neural network (ANN); and obtaining an ANN output vector comprising a detected dispersion matrix index, an antenna index, and a complex-valued symbol drawn from an M-QAM constellation.

[0233] For some embodiments of the further additional example method, the ANN may be configured to have 2 hidden layers.

[0234] Some embodiments of the further additional example method may further include adjusting one or more neurons in at least one level of the ANN to reproduce a target output during a training stage. [0235] For some embodiments of the further additional example method, the received signal may be modulated at a transmitter using MS-STSK symbol transmission.

[0236] For some embodiments of the further additional example method, the received signal may be modulated at a transmitter with an MS-STSK encoder in combination with beam-index modulation, and a vectorized matrix y_BI may be used as the input to the ANN.

[0237] For some embodiments of the further additional example method, the ANN output vector may further include a beam index.

[0238] For some embodiments of the further additional example method, the ANN may include weights and biases determined during a training stage to reduce an error between a target output and a predicted output.

[0239] For some embodiments of the further additional example method, the ANN may include activating functions of a type of hyperbolic tangent function, or a type of sigmoid function.

[0240] For some embodiments of the further additional example method, the received signal may be divided into real and imaginary values by splitting the received vector y into a real part Real(y) and an imaginary part lmag(y) before providing the real part Real(y) as the input to the ANN.

[0241] For some embodiments of the further additional example method, training data for recalibrating the ANN weights may be sent after every N_f frames.

[0242] For some embodiments of the further additional example method, blind-detection may be performed for frames received between training frames.

[0243] Some embodiments of the further additional example method may further include storing the received signal as input samples to a replay buffer.

[0244] A further additional example apparatus in accordance with some embodiments may include: a processor; and a non-transitory computer-readable medium storing instructions operative, when executed by the processor, to: use a vectorized matrix y comprising a received signal of received signal values as an input to an artificial neural network (ANN); and obtain an ANN output vector comprising a detected dispersion matrix index, an antenna index, and a complex-valued symbol drawn from an M-QAM constellation.

[0245] Another example apparatus in accordance with some embodiments may include: an RF combiner; a downconverter; an artificial neural network (ANN) module; and a forward error correction (FEC) decoder;

[0246] Some embodiments of another example apparatus may further include a replay buffer. [0247] Another further example method in accordance with some embodiments may include: establishing a beam alignment between a base station and a user equipment; receiving training data from the base station; training a neural network to detect a data signal from the base station using the training data; receiving a real data transmission signal from the base station; checking a key performance indicator (KPI) with a pre-determined value; and requesting retraining of the neural network if the KPI is less than the predetermined value.

[0248] Another further example apparatus in accordance with some embodiments may include: a processor; and a non-transitory computer-readable medium storing instructions operative, when executed by the processor, to: establish a beam alignment between a base station and a user equipment; receive training data from the base station; train a neural network to detect a data signal from the base station using the training data; receive a real data transmission signal from the base station; check a key performance indicator (KPI) with a pre-determined value; and request retraining of the neural network if the KPI is less than the pre-determined value.

[0249] Note that various hardware elements of one or more of the described embodiments are referred to as “modules” that carry out (i.e., perform, execute, and the like) various functions that are described herein in connection with the respective modules. As used herein, a module includes hardware (e.g., one or more processors, one or more microprocessors, one or more microcontrollers, one or more microchips, one or more application-specific integrated circuits (ASICs), one or more field programmable gate arrays (FPGAs), one or more memory devices) deemed suitable by those of skill in the relevant art for a given implementation. Each described module may also include instructions executable for carrying out the one or more functions described as being carried out by the respective module, and it is noted that those instructions could take the form of or include hardware (i.e., hardwired) instructions, firmware instructions, software instructions, and/or the like, and may be stored in any suitable non-transitory computer-readable medium or media, such as commonly referred to as RAM, ROM, etc.

[0250] Although features and elements are described above in particular combinations, one of ordinary skill in the art will appreciate that each feature or element can be used alone or in any combination with the other features and elements. In addition, the methods described herein may be implemented in a computer program, software, or firmware incorporated in a computer-readable medium for execution by a computer or processor. Examples of computer-readable storage media include, but are not limited to, a read only memory (ROM), a random access memory (RAM), a register, cache memory, semiconductor memory devices, magnetic media such as internal hard disks and removable disks, magneto-optical media, and optical media such as CD-ROM disks, and digital versatile disks (DVDs). A processor in association with software may be used to implement a radio frequency transceiver for use in a WTRU, UE, terminal, base station, RNC, or any host computer.

Claims

1. A method comprising: receiving a wireless signal and performing analog RF combining to generate a combined received signal; down-converting the combined received signal; providing the down-converted combined signal to an artificial neural network (ANN); and using predetermined weights within the ANN to demodulate the received signal in absence of channel state information to obtain data bits.

2. The method of claim 1 , further comprising: recalibrating periodically the predetermined weights using training update information transmitted from the transmitter to the receiver.

3. The method of claim 1, further comprising storing at least one of the data bits or the received, combined, and down-converted signal of received signal values as input samples to a replay buffer.

4. The method of claim 3, wherein the input samples are stored in the replay buffer only if a cyclic redundancy check (CRC) passes.

5. The method of claim 4, further comprising storing a time stamp indicating at least one of when the input samples were collected and how long the input samples will stay in the replay buffer.

6. The method of claim 3, wherein a packet of uncoded bits from the replay buffer is re-encoded to reproduce a transmitter output and corresponding target output of the ANN.

7. The method of claim 3, wherein at least one of: (i) an estimate of a channel coherence time, (ii) a recent bit error rate, or (iii) a packet error rate, is used to determine how long at least one replay buffer entry will remain in the replay buffer.

8. The method of claim 1 , wherein obtaining data bits comprises: inputting to the ANN a vector y, representing a vectorized matrix, comprising the down-converted combined signal of received signal values; and obtaining an ANN output vector comprising a detected dispersion matrix index, an antenna index, and a complex-valued symbol drawn from an M-QAM constellation.

9. The method of claim 8, wherein the ANN is configured to have 2 hidden layers.

10. The method of claim 8, further comprising adjusting one or more neurons in at least one level of the ANN to reproduce a target output during a training stage.

11. The method of claim 8, wherein the received signal was modulated at a transmitter using MS-STSK symbol transmission.

12. The method of claim 8, wherein the received signal was modulated at a transmitter with an MS-STSK encoder in combination with beam index modulation, and wherein the vectorized matrix y is equal to a vectorized matrix y_BI corresponding to beam index modulation.

13. The method of claim 12, wherein the ANN output vector further comprises a beam index.

14. The method of claim 8, wherein the ANN comprises weights and biases determined during a training stage to reduce an error between a target output and a predicted output.

15. The method of claim 8, wherein the ANN comprises activating functions of a type of hyperbolic tangent function, or a type of sigmoid function.

16. The method of claim 8, wherein the received signal is divided into real and imaginary values by splitting the received vector y into a real part Real(y) and an imaginary part lmag(y) before providing the real part Real(y) as the input to the ANN.

17. The method of claim 8, wherein training data for recalibrating the ANN weights is sent after every N_f frames.

18. The method of claim 17, wherein blind-detection is performed for frames received between training frames.

19. The method of claim 8, further comprising storing the received signal as input samples to a replay buffer.

20. An apparatus comprising: a processor; and a non-transitory computer-readable medium storing instructions operative, when executed by the processor, to: receive a signal and perform analog RF combining; down-convert the combined received signal; provide the down-converted combined signal to an artificial neural network (ANN); and demodulate the received signal using predetermined weights in absence of channel state information to obtain uncoded bits.

21. A method comprising: using a vector y, representing a vectorized matrix, comprising a received signal of received signal values as an input to an artificial neural network (ANN); and obtaining an ANN output vector comprising a detected dispersion matrix index, an antenna index, and a complex-valued symbol drawn from an M-QAM constellation.

22. The method of claim 21 , wherein the ANN is configured to have 2 hidden layers.

23. The method of claim 21, further comprising adjusting one or more neurons in at least one level of the

ANN to reproduce a target output during a training stage.

24. The method of claim 21, wherein the received signal was modulated at a transmitter using MS-STSK symbol transmission.

25. The method of claim 21 , wherein the received signal was modulated at a transmitter with an MS-STSK encoder in combination with beam index modulation, and wherein a vectorized matrix y_BI is used as the input to the ANN.

26. The method of claim 25, wherein the ANN output vector further comprises a beam index.

27. The method of claim 21, wherein the ANN comprises weights and biases determined during a training stage to reduce an error between a target output and a predicted output.

28. The method of claim 21 , wherein the ANN comprises activating functions of a type of hyperbolic tangent function, or a type of sigmoid function.

29. The method of claim 21 , wherein the received signal is divided into real and imaginary values by splitting the received vector y into a real part Real(y) and an imaginary part lmag(y) before providing the real part Real(y) as the input to the ANN.

30. The method of claim 21 , wherein training data for recalibrating the ANN weights is sent after every N_f frames.

31. The method of claim 30, wherein blind-detection is performed for frames received between training frames.

32. The method of claim 21, further comprising storing the received signal as input samples to a replay buffer.

33. An apparatus comprising: a processor; and a non-transitory computer-readable medium storing instructions operative, when executed by the processor, to: use a vectorized matrix y comprising a received signal of received signal values as an input to an artificial neural network (ANN); and obtain an ANN output vector comprising a detected dispersion matrix index, an antenna index, and a complex-valued symbol drawn from an M-QAM constellation.

34. An apparatus, comprising: an RF combiner; a downconverter; an artificial neural network (ANN) module; and a forward error correction (FEC) decoder;

35. The apparatus of claim 34, further comprising a replay buffer.

36. A method comprising: establishing a beam alignment between a base station and a user equipment; receiving training data from the base station; training a neural network to detect a data signal from the base station using the training data; receiving a real data transmission signal from the base station; checking a key performance indicator (KPI) with a pre-determined value; and requesting retraining of the neural network if the KPI is less than the pre-determined value.

37. An apparatus comprising: a processor; and a non-transitory computer-readable medium storing instructions operative, when executed by the processor, to: establish a beam alignment between a base station and a user equipment; receive training data from the base station; train a neural network to detect a data signal from the base station using the training data; receive a real data transmission signal from the base station; check a key performance indicator (KPI) with a pre-determined value; and request retraining of the neural network if the KPI is less than the pre-determined value.