CN102421105A - Method for modeling ultra wide-band (UWB) compressive sensing wireless channel - Google Patents

Abstract

The invention discloses a method for processing the channel measurement data of an ultra wide-band (UWB) compressive sensing frequency domain and belongs to the field of communication. The method comprises the following steps of: for UWB channel measurement data in a mode of frequency domain measurement, designing a window function to intercept the channel measurement data of a target frequency range; by using a time domain pulse which corresponds to the window function as prior information, constructing an ultra-complete waveform dictionary to make a time domain channel measurement signal more sparsely represented; by using a greed tracking algorithm as a compressive sensing reconstruction algorithm and reconstructing the time domain channel measurement signal, finishing deconvolution of the signal so as to reduce the sampling rate of measurement equipment on the premise of no reduction of deconvolution accuracy. The invention also discloses a method for modeling a UWB compressive sensing multi-path channel. By using the characteristic of a correspondence relationship between the success reconstruction probability of a group of channel measurement data in a compressive sensing theory and observation sample number, and the characteristic of limitation on the success reconstruction probability by channel average multi-path number, a relationship model between the observation sample number and the channel average multi-path number is established.

Description

Ultra broadband compressed sensing wireless channel modeling method

Technical field:

The present invention is based on compressed sensing (CS, Compressive Sensing) theory, proposed ultra broadband (UWB, Ultra Wide-Band) wireless channel modeling method, belong to the communications field.

Technical background:

The UWB technology is a kind of novel wireless communication technology that folded formula is used frequency spectrum resource that serves as a contrast, and it sends signal bandwidth can be up to number GHz (GHz).According to the UWB spectral mask standard of U.S. FCC and promulgation in 2002, its absolute bandwidth can reach 7.5Ghz.At present; IEEE has issued relevant criterion to the many application scenarioss of UWB; Comprising a high-speed radio accurate IEEE802.15.3 of territory network mark and low-speed wireless sensor network standard IEEE 802.15.4; And the wireless body area network that is used for the medical monitoring field (WBAN, Wireless Body Area Network) of present broad research.On December 6th, 2008; For satisfying the application demand of social every profession and trade to short distance, high data rate communication; Improve the utilance of frequency; China radio control department has issued " the technological frequency of ultra broadband (UWB) is used regulation ", and the available frequency band of dividing for the UWB technology is 6-9GHz and 4.2-4.8GHz (it is auxiliary to need to detect interference avoidance (DAA) technology).Because the very high bandwidth of UWB signal; Thereby its time-domain signal has extremely strong multi-path resolved power; Rake reception technique through the advanced person makes full use of multipath component, in Wireless Personal Network (WPAN, Wireless Personal Local Network), has application potential.In addition, enormous bandwidth causes the UWB signal to possess hi-Fix and range capability, also has the extensive use scene in military field, for example high-resolution radar and imaging system through walls.

UWB technology with numerous technical advantages is considered to the following main flow of short-distance wireless communication technology, and the UWB channel model is the technological bases of all kinds of UWB of assessment.Channel Modeling is the element task of UWB technical research, and the evaluating tool of a justice can be provided for the motion of all kinds of UWB technology, and the design of UWB system, test and improvement all depend on the understanding to channel.When research UWB and other wireless communication system compatibility, the UWB channel also is one of principal element that influences compatibility analysis, is determining the link budget of UWB system.In UWB Study on Technology field, the antenna of employed communication system, the design and analysis of receiver, Communication System Simulation etc. all need an accurate and general channel model.

The Channel Modeling of UWB can be divided into two types: statistics formula model (Statistical Model) and definite formula (Deterministic Model) model.Owing to confirm that the formula channel model needs the characterisitic parameter of body of wall, various barriers to set the environmental parameter of radio wave propagation process, so the more statistics formula channel models that adopt of UWB channel are represented more.Statistics formula channel model is the description to signal wireless communication environments statistical property; Often through combining statistical analysis and experimental technique to obtain; IEEE 802.15.3a and IEEE 802.15.4a channel model all are based on the correction S-V model of Saleh-Valenzuela (S-V) sub-clustering, can be expressed as like IEEE 802.15.4a channel model

$h\left(t\right)=\underset{l=0}{\overset{L}{\mathrm{\Σ}}}\underset{k=0}{\overset{K}{\mathrm{\Σ}}}{a}_{k,l}\exp \left(j{\mathrm{\φ}}_{k,l}\right)\mathrm{\δ}(t-T_l-{\mathrm{\τ}}_{k,l})---\left(1\right)$

Wherein, a _{K, l}The amplitude in k footpath in being l bunch; T _lIt is l bunch the time of advent; τ _{K, l}The time of advent in k footpath in being l bunch; φ _k, _lBe phase place, obey the even distribution in [0,2 π].

Set up statistics formula channel model and at first will measure the propagation characteristic of UWB signal in varying environment, collect a large amount of measurement data.For the UWB Channel Modeling of single input, single output, common frequency domain measurement scheme can reduce: utilize vector network analyzer (VNA) emission simple signal to carry out frequency sweep, obtain the frequency response of a certain frequency range upper signal channel, the measuring system sketch map is as shown in Figure 1.Through reprocessing and parameter Estimation to UWB channel circumstance measurement data, obtain about the time constant UWB channel fading characteristic accurate measurement data.Consider the low characteristics of short haul connection mobility, constant channel is significant during research.

The UWB multipath channel is mainly paid close attention to average multipath number and time delay extended attribute.The main research contents of UWB channel modeling method is post-processing approach and the method for parameter estimation to measurement data.The measurement data reprocessing will be adopted windowing, the technology such as sub-clustering that inverse Fourier transform obtains channel time domain response, CLEAN algorithm deconvolution and discrete channel response will be carried out in the complex frequency response; Adopt parameter Estimation that the multidimensional parameter that comprises multidiameter delay and amplitude distribution is united estimation.The channel model simulation softward of standard is set up and based on the channel test platform of ripe channel simulator in final channel model and the model parameter storehouse thereof with broad applicability that on the basis of extended measurements, form.

Yet, the very high bandwidth of UWB signal (reaching as high as several GHz), if directly the Nyquist sampling is carried out in the UWB pulse, the sampling rate of ADC will reach several GHz even tens GHz at least, over-sampling can further improve the requirement to sampling rate.For the UWB channel measurement, because the very high bandwidth of UWB signal, frequency domain measurement equipment is accomplished a channel measurement often needs the several seconds.Because the frequency sweep time during frequency domain measurement must be less than the coherence time of channel, frequency sweep count too much limited the frequency domain measurement channel method can only the measure static channel.The compressed sensing theory can be used for alleviating the sampling burden, thereby shortens the time of a sweep measurement.The UWB receiver presses for new method and obtains required sample rate and bit resolution.In the UWB channel impulse response, be not that each distinguishable multipath all contains multipath component in the time interval, so the UWB channel is condition of sparse channel, can utilizes CS to solve the too high problem of sample rate.

CS is one of the focus in present signal processing field.Research shows that most signal all is sparse or compressible, and wherein exhausted most information all is not too important, removes these not too important information and only keeps important information information integrity is not had too big infringement.According to the sparse property of signal, only need be lower than a lot of small sample of nyquist sampling rate, just can signal be recovered well.The CS theory is pointed out; Suppose that length is that signal X conversion coefficient on certain group orthogonal basis ψ of N is sparse; If we can find one with transform-based or the basic Φ of the incoherent observation of tight frame ψ: M * N (M＜＜N) signal X is carried out linear transformation; And obtain observing set Y:M * 1, so just can utilize and optimize the method accurate or high probability ground reconstruct primary signal X from the observation set that finds the solution.The main restructing algorithm of CS can be divided into following three major types: protruding method of relaxation, combinational algorithm and greedy tracing algorithm.Greedy tracing algorithm is to select a locally optimal solution progressively to approach primary signal during through each iteration.This type algorithm comprises: match tracing (MP, Matching Pursuit) algorithm, orthogonal matching pursuit (OMP) algorithm, segmentation OMP algorithm (StOMP) and regularization OMP (ROMP) algorithm etc.

This patent provides a kind of new frequency domain channel Measurement and Data Processing method and multipath channel modeling method based on the CS theory.Data processing method principle based on CS is the time domain pulse that utilizes the frequency domain window function corresponding, and the time domain measurement data are carried out rarefaction representation, with CS the time domain channel measuring-signal is carried out deconvolution and obtains the discrete channel response.This method can greatly reduce sample rate under the prerequisite that does not reduce the deconvolution accuracy.Multipath channel modeling method principle based on CS is; On data processed result basis based on CS; Utilize the signal reconstruction probability of CS to receive the characteristics of the sparse property of signal restriction; The random observation of at first the setting up CS relational model with measuring-signal reconstruct probability of counting responds sparse property by average this characteristic of multipath number decision by discrete channel again, sets up the CS observation relational model with the average multipath number of channel of counting.

Summary of the invention:

The present invention provides a kind of UWB wireless channel modeling method based on CS; Can be under the prerequisite that guarantees the Measurement and Data Processing accuracy; Reduce the sample rate of channel measurement receiving system, and stochastical sampling (observation) relational model with the average multipath number of channel of counting is provided.

The present invention provides a kind of UWB frequency domain channel Measurement and Data Processing method based on CS, and its technical scheme is following:

1, to the original UWB channel measurement data under the frequency domain measurement mode, utilize the frequency domain windowing to come intercepting to meet the channel measurement data of target frequency bands;

2, the target frequency bands measurement data after frequency domain window function and the windowing is constructed conjugation symmetry spectrum respectively, carry out inverse Fourier transform, obtain the time domain window pulse and the time domain measurement data of real number value;

3, utilize the corresponding time domain pulse of frequency domain window function to construct dictionary, the time domain measurement data are carried out rarefaction representation as prior information;

4, CS adopts greedy restructing algorithm, when measurement data is carried out reconstruct, accomplishes the deconvolution to the time domain measurement data, obtains the discrete channel response and estimates;

5, change the dictionary building method that is made up of the time domain window pulse, the discrete channel response that can obtain different resolution is flexibly estimated.

The present invention also provides a kind of average multipath number of UWB channel model based on CS, and its technical scheme is following:

1,, sets the complete reconstruct thresholding of CS of acceptable individual channel measuring-signal according to the channel response of reconstruct and the energy error of real channel measurement;

2, the time domain measurement data of target frequency bands are classified by measurement environment, comprise and distinguish scene type, sighting distance (LOS) or non line of sight (NLOS) and channel measurement distance range etc.;

3, respectively every group of channel measurement data are carried out stochastical sampling by channel circumstance, utilize CS to carry out every group of channel measurement data reconstruction, change the random observation value number of CS, obtain random observation value number and the relation curve of reconstruct probability under the varying environment;

4, utilize the signal reconstruction probability of CS determined by average multipath number by the characteristics of the sparse degree restriction of signal, sparse degree, thereby set up the relational model of CS random observation value number and the average multipath number of channel.

The present invention has following beneficial effect:

When 1, the UWB channel measurement of technical scheme of the present invention under being used for the frequency domain mode handled, can under the prerequisite that does not reduce the deconvolution accuracy, greatly reduce the sample rate of channel measurement equipment.

2, technical scheme of the present invention has been set up and has been observed the relational model of counting with the average multipath number of channel, can put under the said conditions at fixed observer, obtains the average multipath number of channel according to the reconstruct probability of one group of measuring-signal.

3, technical scheme of the present invention realizes flexibly, can obtain the discrete channel response of different resolution according to the parameter in the adjustment parametrization waveform dictionary.

Description of drawings:

Fig. 1 is the measuring system block diagram under the frequency domain measurement mode.

Fig. 2 is the frequency domain window function with Gaussian window and Kai Ze window transition band characteristic.

Fig. 3 type of a being Gaussian window and type corresponding time domain window pulse of triumphant damp window.

Fig. 4 (a) is the contrast sketch map of office scenarios LOS situation next actual measurement channel response and reconstruction signal; Fig. 4 (b) is the discrete channel response of this actual measurement channel response after based on the CS deconvolution.

Fig. 5 is under the office scenarios LOS situation, CS-MP and the energy capture rate simulated effect figure of CLEAN algorithm under different iterationses.

Fig. 6 is under the office scenarios NLOS situation, CS-MP and the energy capture rate simulated effect figure of CLEAN under different iterationses.

Fig. 7 is under office scenarios LOS and the NLOS situation, the observation of the CS relational model analogous diagram with successful reconstruct probability of counting.

Embodiment:

In order to make the object of the invention, technical scheme and advantage clearer, below in conjunction with accompanying drawing and execution mode, be example with the UWB Channel Modeling that meets Chinese UWB spectrum criterion, the present invention is further elaborated.Only should be pointed out that in order to explaining the present invention, and be not used in qualification the present invention.

Please refer to Fig. 1, be the UWB channel frequency domain measurement system block diagram of the present invention's employing.The embodiment of the invention has been measured polytype environment such as office, dwelling house, outdoor open ground, and measurement data is divided into LOS and two kinds of situation of NLOS.Measurement parameter is for measuring frequency range 2.3-11GHz, and emission simple signal number is 5600, and frequency sweep is spaced apart 1.5536MHz, transmitting power 10dBm, and measuring range is 1-10m.

In order to obtain meeting the channel transfer functions of Chinese spectrum criterion, need carry out the measuring-signal of frequency domain windowing intercepting 6～9GHz frequency range to the original measurement signal

Y(f)＝H(f)W(f) (2)

Wherein Y (f) is the channel transfer functions after the windowing, and H (f) is the channel transfer functions of measuring, and W (f) is the frequency domain window function.

If directly the data of target frequency bands are done inverse Fourier transform, are equivalent to H (f) is added the rectangular window of target frequency bands.Adopt the window function of which kind of form can have influence on channel measurement Data Post based on CS.Descend because the time delay secondary lobe of rectangular window is 1/t reciprocal in time, can occur the so-called mediation conditions of streaking of crossing in the time domain impulse response, cause estimated channel impulse response (CIR) that the expansion of bigger root mean square (RMS) time delay is arranged.If but the transition band of window is limited in the 6-9GHz frequency range, can reduce the bandwidth of channel frequency response again, reduce temporal resolution.Because institute's frequency measurement band broad can design the transition band of window outside target frequency bands.The time domain pulse that Gaussian window is corresponding be Gauss's form still, and secondary lobe is less, and Gaussian window has good time-frequency aggregation, so the present invention adopted type Gaussian window of transition band for Gauss's roll-off characteristic, and its frequency domain representation does

$W\left(f\right)=\left\{\begin{array}{cc}\frac{1}{a}{e}^{-\frac{{(f-6)}^2}{b}},& f\∈\left(\mathrm{5,6}\right)\\ 1,& f\∈\left[\mathrm{6,9}\right]\\ \frac{1}{a}{e}^{-\frac{{(f-9)}^2}{b}},& f\∈\left(\mathrm{9,10}\right]\\ 0,& f\∈\left(\mathrm{10,11}\right)\end{array}\right.---\left(3\right)$

Wherein a and b represent the coefficient of transition band roll-off characteristic, and the unit of f is GHz.After the 10-11GHz zero padding, the frequency spectrum digital bandwidth that W (f) is corresponding has reached 6GHz, and temporal resolution is 0.167ns.

By formula (3) structure type Gaussian window, when a=1 and b=0.093, the three dB bandwidth and the 20dB bandwidth of type Gaussian window are respectively 3.51GHz and 4.31GHz, the only about 650MHz of its transition band width.Please refer to Fig. 2, be class Gaussian window with Gauss's roll-off characteristic transition band and the triumphant damp window of class with triumphant damp window roll-off characteristic transition band, wherein the parameter of type triumphant damp window is β=6.5, and the 20dB bandwidth is 4.7GHz.Please refer to Fig. 3, be the time-domain representation form of class Gaussian window, type triumphant damp window and 6-9GHz rectangular window.As can beappreciated from fig. 3, the secondary lobe error ratio of rectangular window is slower, and conditions of streaking is serious.The class Gaussian window is less with the secondary lobe of type triumphant damp window, and waveform is more similar.

In order to obtain real-valued time domain measurement signal, the complex frequency response that VNA is exported is configured to conjugation symmetry spectrum.The channel measurement signal of time domain is that conjugation symmetry spectrum is used the result that inverse Fourier transform obtains

y(t)＝TF ^-1[Y(f+f _c)+Y ^*(-f+f _c)] (4)

f _cFor Y (f) is moved on to the required frequency variable of base band, get f here _c=5GHz, TF ^-1The expression inverse Fourier transform.The time domain pulse that window function is corresponding does

s(t)＝TF ^-1[W(f+f _c)+W ^*(-f+f _c)] (5)

Time-domain response through obtaining after the inverse Fourier transform shown in the formula (4) is not the Dirac impulse response that formula (1) is described, and this blocks mainly due to windowing is carried out in frequency response, and pulse s (t) secondary lobe that window function is corresponding superposes.Y (t) is the channel impulse response of formula (1) expression and the convolution of basic waveform s (t)

$y\left(t\right)=s\left(t\right)\⊗\underset{i=l}{\overset{L}{\∑}}{\mathrm{\α}}_i\mathrm{\δ}\left(\mathrm{\τ}{-\mathrm{\τ}}_i\right)+n_w\left(t\right)---\left(6\right)$

N wherein _w(t) be remaining white Gaussian noise after the windowing.

In order to obtain CIR h (t), formula (6) needs a deconvolution algorithm.The CLEAN algorithm is to use the high-resolution deconvolution algorithm that arrives always in the UWB Channel Modeling in the past, has than Fourier analysis higher accuracy is arranged.But the CLEAN algorithm needs a lot of measurement points of frequency sweep of VNA.The CIR of UWB is that the time is sparse, promptly is not all to comprise multipath component in each distinguishable time delay spacing, and this phenomenon is the fundamental characteristics of UWB channel, and is relevant with the distribution of scatterer in the channel.Therefore the UWB channel is sparse under many scenes, and the sparse property of UWB channel makes the application of CS become possibility.

Directly on the sparse model of time domain to UWB signal application CS, the reconstruct poor-performing is because noise can reduce the sparse property of time domain that receives signal.With time-domain signal on some base or dictionary, launch can enhancing signal sparse property.Designing the sparse property that a parametrization waveform dictionary can strengthen the time domain measured signal representes.Among the present invention, owing to time domain UWB channel measurement signal y (t) is made up of the window pulse s (t) of different gains, time delay, dictionary should be by the atomic building of ability perfect representation UWB measuring-signal.In order to be closely related with the window pulse waveform, the different time shift versions of the pulse s (t) that the dictionary atom should be provided by formula (5) constitute, and are promptly generated by parameterized waveform

d _i(t)＝s(t-iΔ)，i＝0，1，2，... (7)

Parametrization waveform dictionary is defined as

D＝{d ₀(t)，d ₁(t)，d ₂(t)，...} (8)

If y is the discrete representation of y (t) in the formula (4)

y＝[y(0)，y(T _s)，...，y((N-1)T _s)] ^T (9)

T wherein _sBe the temporal resolution of y (t) in the formula (4), N is counting of time-domain response, and T is the transposition symbol.

Formula (7) by continuous time t and Δ explain, two parameters should be discretizations in the reality, the minimum delay unit of the corresponding multipath of atom step delta can be set to the integral multiple of temporal resolution, the Δ of different value will obtain the CIR of different multi-path resolved rates.Change Δ, can design the waveform dictionary of various multi-path resolved rates neatly.With the atom uniform sampling among the dictionary D, obtain discrete dictionary

Ψ＝{d ₀，d ₁，d ₂，...} (10)

So obtain

y＝ΨΘ+n _w (11)

N wherein _wIt is the additive white Gaussian noise (AWGN) of discrete form.Formula (11) shows that receiving signal y has had compacter expression-form.

It is K * N random matrix that matrix Φ is measured in definition, and matrix element is obeyed the normal distribution of zero-mean, unit variance, i.e. φ _{I, j}～N (0,1).Make g=Φ that y is the accidental projection signal, obtain incoherent measured value

g＝ΦΨΘ+Φn _w (12)

Wherein Θ is sparse vector Θ=[θ ₁, θ ₂..., θ _Z] ^T, Z is the atom number among the dictionary Ψ.Make V=Φ Ψ={ v ₁, v ₂..., v _Z, V is called holographic dictionary (holographic dictionary).Suppose that y is that the M rank are sparse on Ψ, measure matrix Φ for k * N, the CS theory shows, has oversample factor c＞1, and the feasible individual incoherent measured value of K=cM that only needs promptly can high probability reconstruct y.On accidental projection signal g and holographic dictionary V, use l ₁The norm minimization algorithm is come reconstruction signal.For formula (12), when noise was very little, Θ was the unique solution of following protruding optimization problem with very high probability

min||Θ|| ₁，s.t.g＝VΘ (13)

Because y is the signal of a noisy, accurately reconstruct is impossible.Yet UWB channel measurement often distance is short, and the transmitting power setting is bigger, can think that noise effect is less.

In the CS theory, the reconstruct probability both can refer to the probability of CS to the accurate reconstruct of individual signals, can refer to again under the fixed sample rate condition, and when mass data (signal) is carried out reconstruct, the ratio of data in total data of accurate reconstruct.Because statistics formula Channel Modeling need be handled mass data, so the embodiment of the invention adopts back a kind of definition mode of reconstruct probability.

The CS restructing algorithm adopts greedy tracing algorithm, and the embodiment of the invention adopts the MP algorithm as an example.In the iterative process of MP, the dictionary of utilization and g close match finds the atom that links together with multipath successively by multipath gain size, and obtains the estimation of channel impulse response, thereby accomplishes deconvolution.The MP algorithm can be described below

Step 1 initialization: signal residual error e ₀=y; Estimated result does

Iteration count t=1;

Step 2 selects to be matched with most in the holographic dictionary atom of signal residual error:

Step 3 update signal residual values is also made estimation to the coefficient of selected atom:

$e_t=e_{t-1}-\frac{|<e_{t-1,v_{l,}}>|}{\left|\right|v_{e,}{\left|\right|}^2}{v}_{l,},$ ${\widehat{\mathrm{\&theta;}}}_{l_t}={\widehat{\mathrm{\&theta;}}}_{l_t}+\frac{|<e_{t-1,}{v}_{l_t}>|}{\left|\right|{v_l}_t{\left|\right|}^2};$

Step 4 inspection convergence: if t＜T ₀And || e _t|| ₂＞ε || y|| ₂, make t=t+1 change step 2; Otherwise change step 5;

The estimated result of step 5 reconstruction signal is:

The MP algorithm is exported the gain that a sparse estimator

is wherein comprising each multipath in the signal reconstruction process.Particularly as Δ=T _sThe time, Z=N is arranged, the CIR that this moment, deconvolution obtained has highest resolution.In fact the MP algorithm is to be applied on the g of projection signal that uses holographic dictionary V, yet its effect can be better understood on primary signal y and dictionary Ψ.

Traditional CLEAN algorithm utilizes the correlation of signal, carries out through the several times iteration.The CLEAN algorithm pollutes relevant between drawing (dirty map) prior information (template) through serial deletion, reconstructs clean drawing (clean map), wherein pollutes the measuring-signal that drawing promptly receives noise pollution, and clean drawing refers to the CIR that estimates.Make the corresponding time domain form s (t) of window function be the template of CLEAN, i.e. p (t)=s (t).The step of CLEAN algorithm is described below

Step 1 is with dirty drawing e ₀(t) with being initialized as e ₀(t)=y (t), and clean drawing is initialized as c ₀(t)=0;

Step 2 calculating normalization cross-correlation function R _Ep(τ)=e _N-1(t) ⊙ p (t) calculates

With

(⊙ representes related operation);

The dirty drawing of step 3 cleaning, also is updated to

with clean drawing

If all

of step 4 changes step 5; Otherwise, change step 2;

Step 5 channel impulse response estimation is

Can find out that CLEAN algorithm and MP algorithm are closely similar.Contrast two algorithms, the step 2 that can find MP is dictionary atoms that selection and residual error are mated most

And the step 2 of CLEAN is to find out the maximum of cross-correlation function

The 1st atom supposing dictionary Ψ is d ₀=[s ₁s ₂... s _P0...0] ^TAnd s=[s ₁s ₂... s _P] ^T, P is the vector length of s.If Δ is set to temporal resolution T _s, Ψ can be expressed as so

Consider

And d _i(t)=s (t-i Δ), N＞＞during P, Ψ is arranged _TΘ ≈ s ⊙ Θ, promptly step 2 essence of two algorithms is identical.Therefore, MP algorithm and CLEAN algorithm are of equal value.Yet MP is the restructing algorithm of CS, uses holographic dictionary V, and application is accidental projection signal g, is not to use dictionary Ψ to be applied on the measuring-signal y, and this is the main distinction of CS-MP and CLEAN.

Utilize the deconvolution performance of frequency domain channel measurement data check CS-MP, and compare to identical windowing time-domain signal application CS-MP and CLEAN.The performance of deconvolution can be represented with respect to the energy capture rate of original measurement signal with reconstruction signal.

Please refer to Fig. 4, be the deconvolution simulation result of one under office scenarios LOS situation actual measurement channel response based on the CS-MP algorithm.The iterations of MP is made as T ₀=600, Fig. 4 (a) is the waveform of original measurement signal and with the contrast sketch map of CS-MP reconstruction signal, and its energy reconstructed error is 0.047; Fig. 4 (b) is the channel impulse response after the CS-MP deconvolution.

Please refer to Fig. 5-6, be respectively under office scenarios LOS and the NLOS situation, CS-MP and the energy capture rate curve analogous diagram of CLEAN under different iterationses.CS-MP curve among Fig. 5 has comprised a type Gaussian window, two kinds of situation of rectangular window, and wherein under type Gaussian window situation, measurement point number K/N of CS-MP gets 0.5,0.6 and 0.7 successively.Can draw from Fig. 5 and Fig. 6, can use less measured value to obtain the deconvolution performance close with the CLEAN algorithm based on the deconvolution algorithm of compressed sensing.The energy capture rate of two algorithms all rises along with the rising of iterations.CS-MP also can obtain higher deconvolution performance with the increase of population of measured values.Yet CS-MP is after iterations reaches certain value, and the energy capture rate almost no longer rises.Y is that the M rank are sparse on Ψ if this is, has oversample factor c＞1 among the CS, needs K=cM incoherent measured value ability high probability reconstruct y at least.Yet it is not desirable sparse that measuring-signal receives The noise, and when sampling number K was certain value and K＜cM, its reconstructed error can be not infinitely small, but tend to a fixed value.

Fig. 5 also provides the deconvolution performance of CS-MP when using rectangular window.Can find out, when using the class Gaussian window, under same sample rate K/N=0.5 condition, use the energy capture rate of type Gaussian window to be better than the energy capture rate of using rectangular window.Because the time domain waveform side lobe attenuation that the frequency domain rectangular window is corresponding is slow, the stack of its secondary lobe causes the sparse property of time domain measurement data to descend to some extent, reduces the deconvolution accuracy of CS.

Investigate UWB channel time domain measuring-signal y with CS, the average multipath number model of setting up under the CS framework adopts following technical scheme:

UWB channel measured data is divided into groups by scene type, LOS or factors such as NLOS, measuring distance.For example, the present invention is divided into two groups with the measurement data in the office scenarios by LOS and NLOS, carries out reconstruct with CS, and restructing algorithm adopts the MP algorithm, and other conditions are constant.There is average 10dB multipath number parameter N in every group of measuring-signal _10dB, definition 10dB multipath number is that the maximum diameter of a measured signal is decayed 10dB with interior multipath number.

The observation that constantly changes CS is counted, and writes down this observation measured data reconstruct probability down of counting, and the CS that then obtain many group measured datas observe and counting and reconstruct probability corresponding relation.For the individual channel measuring-signal; The energy error that success reconstruct is defined as reconstruction signal and actual signal is in 10%; Promptly

defines the ratio of successful reconstruct probability for the number and the total number of this group measuring-signal of successful reconstruct in this group channel measurement signal for every group of channel measurement signal.

Please refer to Fig. 7, be the successful reconstruct probability of actual measurement channel and the relation of sampling number under office scenarios LOS and two kinds of situation of NLOS.Each bar curve is all represented the average multipath number that a measured data is divided into groups among Fig. 7, counts and the average multipath number of UWB channel relational model thereby obtain CS observation.According to Fig. 7,, can find out that will make the reconfigurable measurement signal will reach 90% successful reconstruct probability, the LOS of office situation needs about 500 random observation points, corresponding N through the observation of the CS corresponding relation with the reconstruct probability of counting _10dB=78; The NLOS of office situation needs about 850 random observation points, corresponding N _10dB=105.

Please refer to Fig. 7, reach identical reconstruct probability, compare the LOS scene, the NLOS scene needs more stochastical sampling to count.The following fact of this phenomenon is consistent: more refracted, reflection etc. have taken place in the UWB signal under the NLOS scene, so than having more multipath component under the LOS scene.Therefore, the measuring-signal of LOS situation is more sparse than the measuring-signal under the NLOS situation under the scene of the same race, needs sampling number still less during reconstruct.Please refer to Fig. 5-6, can find out from the contrast of two width of cloth figure, under identical sample rate and iterations, CS-MP is better than the NLOS data to the deconvolution performance of LOS data, and reason is that the LOS data are stronger than the sparse property of NLOS data equally.

Claims (6)

1. ultra broadband compressed sensing wireless channel modeling implementation method comprises:

Utilize frequency domain windowing mode intercepting target frequency bands to operate and obtain the channel measurement data;

Based on the UWB channel frequency domain measurement data processing operation of compressed sensing, be used to estimate the discrete channel response;

And the observation of setting up CS is counted and the average multipath number of UWB channel relational model.

2. method according to claim 1; It is characterized in that; Said utilize frequency domain windowing mode intercepting target frequency bands to operate to obtain the channel measurement data comprise: under UWB channel frequency domain measurement mode, the actual measurement frequency range is greater than target frequency bands, with the transition band design of window function outside target frequency bands; The amplitude-frequency characteristic of window function in target frequency bands is smooth straight line, and the transition band outside target frequency bands is a smooth curve.

3. method according to claim 1; It is characterized in that; Said UWB frequency domain measurement data processing operation based on compressed sensing comprises: utilize the corresponding time domain window pulse constructing variable waveform dictionary of frequency domain window function; Strengthen the sparse expression of time domain measurement signal, utilize compressed sensing that the time domain measurement signal is carried out deconvolution, obtain the discrete channel response and estimate.

4. method according to claim 3; It is characterized in that; The said corresponding time domain window pulse constructing variable waveform dictionary of frequency domain window function that utilizes comprises: window pulse arranged according to equally spaced mode, and by arrangement pitch parameter determining compressed sensing deconvolution result's multipath minimum interval.

5. method according to claim 3; It is characterized in that; The sparse expression of said enhancing time domain measurement signal, utilize the compressed sensing deconvolution to obtain the discrete channel response and estimate to comprise: the time domain measurement signal indication that the frequency domain channel measuring-signal is corresponding is the convolution form of time domain window pulse and discrete channel response, adopts the restructing algorithm of greedy tracing algorithm as compressed sensing; Make compressed sensing in signal reconstruction, accomplish deconvolution, obtain the discrete channel response and estimate.

6. method according to claim 1; It is characterized in that; The relational model that said observation of setting up compressed sensing is counted with the average multipath number of UWB channel comprises: to the channel measured data according to scene type, factor such as sighting distance, measuring distance scope is classified; Utilize compressed sensing to carry out reconstruct to the time domain measurement signal; The observation that constantly changes CS is counted, and writes down this observation measured data reconstruct probability down of counting, and the CS that obtain many group measured datas observe and counting and reconstruct probability corresponding relation curve; The average multipath number that all corresponding measured data of each bar curve is divided into groups, the observation that obtains CS is counted and the average multipath number of UWB channel relational model.

Images