CN101247376A - OFDM channel estimation method for eliminating noise combined with wavelet transformed domain - Google Patents

Abstract

The invention discloses an OFDM channel estimation method by combining wavelet transform domain denoise method, comprising the following steps of: extracting the information on a pilot subcarrier, calculating the channel frequency response where the pilot subcarrier is; carrying out interpretation to the channel frequency response in the direction of time according to the calculated channel frequency response on the location of the pilot; the channel frequency response of the interpretation on the direction of time is removed noise and interference; carrying out the interpretation on the direction of frequency to the channel frequency response removed noise and interference on the direction of frequency channel. Because the noise, interference between subcarriers and linear interpolation can bring errors into the channel frequency response obtained by the interpretation, the errors can reduce the performance of the OFDM(Orthogonal frequency-division multiplexing) receiver. The method based on the wavelet transform domain is adopted to reduce the errors in the channel frequency response, increase the accuracy of the estimated value of the channel, and reduce the error rate of receiving system.

Description

The OFDM channel estimation methods of eliminating noise combined with wavelet transformed domain

Technical field

The present invention relates to radio broadcasting system, particularly the OFDM of eliminating noise combined with wavelet transformed domain (OFDM) channel estimation methods.

Background technology

DRM (digital radio globalization) is unique universal non-patent digital radio broadcasting system of shortwave, medium wave and long wave amplitude modulation broadcasting frequency range.Under same coverage condition, DRM (digital radio globalization) transmitter power is than the low 6-9dB of traditional analog transmissions acc power, and digital broadcasting is lower than the same adjacent frequency protective rate of analog broadcasting, and the anti-multipath interference performance is strong, is convenient to move receive; Tonequality can reach the quality of CD (CD) or FM multiplex; Additional data and multimedia messages can be provided; Compare with DAB (digital audio broadcasting), its receiver price is easier to be accepted by the mass audience.Its appearance is the sign that the following frequency range broadcasting of 30MHz is revived, and has become international standard at present.

In coherent demodulation OFDM (OFDM) system, in order to carry out equilibrium to the received signal, receiver must obtain the amplitude and the phase information of channel by channel estimating.But, broadcast channel not only suffers because the frequency selective fading that multipath transmisstion causes, and the time selective fading that brought by Doppler frequency shift or Doppler's expansion, for the quality of reception that guarantees receiver and the real-time of reception, require receiver that broadcast channel is carried out promptly and accurately channel estimating.According to the requirement of DRM (digital radio globalization) standard, transmitter also sends pilot data simultaneously when sending useful data, so just can adopt the channel estimation scheme based on pilot tone.At first, extract the received signal of pilot frequency locations, utilize the local pilot tone of receiver storage, calculate the channel frequency response of pilot frequency locations with least-squares algorithm, then, estimate the channel frequency response at data subcarrier place with interpolation filter, last, with the frequency-domain equalizer of single tap the reception receipt is carried out equilibrium.When sending signal employing high order modulation,,, need channel estimating more accurately in order to obtain better receiver performance such as 16QAM (16 constellation point quadrature amplitude modulation) or 64QAM (64 constellation point quadrature amplitude modulation).

Because under two fading channel conditions, just existing by many frequency selective fadings that cause through propagation, have again because under the channel condition of the time selective fading that Doppler frequency shift or Doppler's expansion cause, the pilot frequency design of rhombus has better anti-fading characteristic than block pilot frequency design or dressing pilot frequency design, just adopted the rhombus pilot frequency design of this time-frequency two-dimensional in DRM (digital radio globalization) standard, this scheme can reduce the degree that is subjected to receiver decreased performance under the situation about having a strong impact on that channel causes in some pilot tone.Rhombus pilot frequency design distribution map as shown in Figure 1.

At different channel conditions, comprised four kinds of different robust modes in DRM (digital radio globalization) standard, concrete description sees Table 1:

Table 1: robust mode and corresponding channel condition

Robust mode	Channel condition
		Mode A	Gaussian channel, the slight rate channel that falls is applicable to the medium wave and the long wave channel on daytime.
Mode B	Time and frequency-selective channel are than the shortwave and the medium wave channel at night of long delay expansion.
		Pattern C	Time and frequency-selective channel, channel condition is relatively poor, the short wave channel of bigger Doppler's expansion.
Pattern D	Unusual Jian Zhuan pattern, but because pilot interval has too closely influenced message transmission rate.

Different robust modes all has different pilot intervals on time orientation and frequency direction, concrete at interval big or small as shown in table 2:

Table 2: pilot interval size

Robust mode	N T	N F
			A	5	20
B	3	6
			C	2	4
D	3	3

In table 2, N _TAnd N _FPilot interval on the difference express time direction and the pilot interval on the frequency direction.

First three plants the application that robust mode can satisfy most of DRM (digital radio globalization) broadcasting, for Mode A, because short protection interval and narrower make it not be suitable for shortwave broadcasting at subcarrier spacing.Have only pattern D to be applicable to that channel model 6 in the standard, this channel model not only have very long time delay expansion, also have very big Doppler's expansion, it is a kind of approximate simulation to the shortwave propagation of region of the equator.

As everyone knows, optimum channel estimator under the minimum mean square error criterion is two-dimentional Weiner filter, but two-dimentional Weiner filter is not easy to realize in practical engineering application very much, but when channel is the steady irrelevant scatter channel of broad sense, the one-dimensional filtering device of two cascades is a kind of good selection schemes, frequency direction interpolation after can first time orientation interpolation, also can first frequency direction interpolation after the time orientation interpolation.We have provided the block diagram (frequency direction behind the first time orientation) of the one dimension interpolation filter of two cascades in Fig. 2:

For OFDM (OFDM) system of assisting based on pilot tone, common channel estimation methods is, at first obtain the channel frequency response of pilot frequency locations with least-squares algorithm, adopt two cascade one-dimensional filtering devices that the channel frequency response of data position is estimated then, for example: at first carry out simple linear interpolation at time orientation, and then in the enterprising line linearity interpolation of frequency direction.

How to improve the performance that all adopts the channel estimation methods of linear interpolation on the time-frequency two-dimensional direction based on pilot tone.FOR ALL WE KNOW, owing to interference and linear interpolation between noise, subcarrier can be introduced error in the channel frequency response value that we obtain by interpolation at last, these errors can cause the decline of OFDM (OFDM) receiver performance.

Summary of the invention

The OFDM channel estimation methods that the purpose of this invention is to provide a kind of eliminating noise combined with wavelet transformed domain.

For achieving the above object, a kind of OFDM channel estimation methods of eliminating noise combined with wavelet transformed domain comprises step:

A) information on the extraction pilot sub-carrier, the channel frequency response at calculating pilot sub-carrier place;

B) according to the channel frequency response at the pilot frequency locations place that calculates, channel frequency response is carried out interpolation at time orientation;

C) channel frequency response to the time orientation interpolation carries out noise and disturbs removal;

D) channel frequency response of having removed noise and interference is carried out the interpolation of frequency direction in the channel direction.

Because interference and linear interpolation can be introduced error between noise, subcarrier in the channel frequency response value that we obtain by interpolation at last, these errors can cause the decline of OFDM (OFDM) receiver performance.And adopt above-mentioned method can reduce the error of introducing in the channel frequency response value based on the wavelet transformed domain denoising, and improve the accuracy of channel estimation value, reduce the error rate of receiving system.

Description of drawings

Fig. 1 is a rhombus pilot frequency design distribution map;

Fig. 2 is two cascade one dimension interpolation filter block diagrams;

Fig. 3 is the cascade interpolation method block diagram of band noise remove;

Fig. 4 is a noise remove process block diagram;

Fig. 5 is the wavelet decomposition block diagram;

Fig. 6 is the wavelet reconstruction block diagram;

Fig. 7 is the bit error rate characteristic of channel 3;

Fig. 8 is the bit error rate characteristic of channel 4.

Embodiment

In order to eliminate the influence of aforesaid noise and interference, after our interpolation on time orientation, increased a noise and disturbed the process of removing, we carry out the interpolation on the frequency direction more then, and its block diagram is as shown in Figure 3.Because interference and linear interpolation can be introduced error between noise, subcarrier in the channel frequency response value that we obtain by interpolation at last, these errors can cause the decline of OFDM (OFDM) receiver performance.The present invention has introduced a noise remove process between time orientation interpolation device and frequency direction interpolation device, its basic principle is by wavelet transformation the channel frequency response that the time orientation interpolation obtains to be carried out noise remove, thereby improve the accuracy that is used for the frequency direction interpolation, get to and reduce channel estimation errors, reduce the purpose of the receiver error rate under the identical signal to noise ratio.Noise remove process among Fig. 3 as shown in Figure 4.

After channel frequency response at the pilot sub-carrier place carried out inverse Fourier transform, we had just obtained the sample value sequence of time domain channel impulse response.We can notice that except useful channel magnitude and phase information, each sample point has also comprised the interference of noise and adjacent sub-carrier simultaneously.These interference and noise can cause channel frequency response that we obtain by interpolation and the error between its actual value.

In order to improve the performance of DRM (digital radio globalization) receiver, we carry out separating of useful channel information and noise and interference component with the discrete time wavelet transformation.In fact, the discrete time wavelet transformation is equivalent to a bank of filters, comprise high pass filter and low pass filter, the output sequence after list entries and the low pass filter convolution is known as scale coefficient, and the output after list entries and the high pass filter convolution then is known as wavelet coefficient.If wavelet coefficient is carried out threshold process, and keep scale coefficient constant, and then by the reconfigurable filter group to being reconstructed through the wavelet coefficient and the untreated scale coefficient of threshold process, our just can be eliminated list entries of noise and interference.At last, we carry out Fourier transform to the list entries of reconstruct again, obtain the channel impulse response on each pilot sub-carrier after the denoising, thereby reach the purpose that improves precision of channel estimation.

Top noise and interference removal process are made up of five steps, specifically describe as follows:

1) inverse Fourier transform

The sequence that channel frequency response on the pilot sub-carrier constitutes is carried out inverse Fourier transform, obtains the channel impulse response sample value sequence on the time domain:

$h_{n,l}=\frac{1}{M}\underset{k=0}{\overset{M-1}{\mathrm{\Σ}}}{\hat{H}}_{\mathrm{ls}}^{k,l}{e}^{j2\mathrm{\πkn}/M},0\≤n\≤M-1---\left(1\right)$

Wherein, M is the number of pilot sub-carrier in an OFDM (OFDM) symbol.

2) wavelet decomposition

Here, we decompose and reconstruct with the Harr small echo, are made up of a high pass and a low pass based on the bank of filters of Harr small echo, and its low-pass filter coefficients is $f_L=\left[\begin{array}{cc}\frac{\sqrt{2}}{2}& \frac{\sqrt{2}}{2}\end{array}\right],$ The high pass filter coefficient is $f_H=\left[\begin{array}{cc}-\frac{\sqrt{2}}{2}& \frac{\sqrt{2}}{2}\end{array}\right].$ The main purpose of this step is exactly to have the channel impulse response sample value sequence transformation of noise and interference to wavelet field, if use h _{N, l} ^LAnd h _{N, l} ^HRepresent scale coefficient and wavelet coefficient behind the wavelet transformation respectively, the mathematic(al) representation of wavelet decomposition can followingly be represented so:

$h_{k,l}^L=\underset{m=0}{\overset{1}{\mathrm{\&Sigma;}}}{h}_{2k-m,l}{f}_L\left(m\right)$ $0\&le;k<\frac{M}{2}---\left(2\right)$

$h_{k,l}^H=\underset{m=0}{\overset{1}{\mathrm{\&Sigma;}}}{h}_{2k-m,l}{f}_H\left(m\right)$ $0\&le;k<\frac{M}{2}---\left(3\right)$

Wherein, h _{K, l}, 0≤k＜M represents channel impulse response sample value sequence, M is the length that needs the channel impulse response sample value sequence of processing.

Wavelet decomposition is as shown in Figure 5:

3) threshold process

The main purpose of this process is exactly that the wavelet coefficient that will be lower than some threshold values all is changed to zero.Because in the wavelet coefficient after the wavelet decomposition, what bigger wavelet coefficient comprised is useful information, and mainly be noise and interference component in the less wavelet coefficient, so we can be changed to the wavelet coefficient that is lower than certain threshold value zero, thereby reach the purpose of removing noise and interference.

Threshold method commonly used mainly contains three kinds, and they are hard-threshold method, soft-threshold method and non-negative threshold value method, and it is defined as follows:

${\stackrel{\&OverBar;}{h}}_{k,l}^H=\left\{\begin{array}{cc}0& \left|h_{k,l}^H\right|\&le;\mathrm{\&lambda;}\\ h_{k,l}^H& \left|h_{k,l}^H\right|>\mathrm{\&lambda;}\end{array}\right.$ (hard-threshold) (4)

${\stackrel{\&OverBar;}{h}}_{k,l}^H=\left\{\begin{array}{cc}0& \left|h_{k,l}^H\right|\&le;\mathrm{\&lambda;}\\ h_{k,l}^H-\mathrm{\&lambda;}& h_{k,l}^H>\mathrm{\&lambda;}\\ h_{k,l}^H+\mathrm{\&lambda;}& h_{k,l}^H<-\mathrm{\&lambda;}\end{array}\right.$ (soft-threshold) (5)

${\stackrel{\&OverBar;}{h}}_{k,l}^H=\left\{\begin{array}{cc}0& \left|h_{k,l}^H\right|\&le;\mathrm{\&lambda;}\\ h_{k,l}^H-{\mathrm{\&lambda;}}^2/h_{k,l}^H& \left|h_{k,l}^H\right|>\mathrm{\&lambda;}\end{array}\right.$ (non-negative threshold value) (6)

Wherein, λ represents threshold value, and λ 〉=0.

The method that wavelet transformed domain is removed noise and interference is different with traditional denoising method based on low pass filter, owing to adopted threshold process process to wavelet coefficient, so it is non-linear.Wherein, choosing of threshold value is very important, and generally, selecting a suitable threshold is an important prerequisite that guarantees good noise and disturb removal effect.If selection of threshold gets too big, some important useful information will be by filtering so, otherwise, too little if selection of threshold gets, when we are reconstructed, will comprise many interference and noise contribution, thereby not reach noise and disturb the purpose of removing.The empirical equation of the wavelet threshold that Donoho proposes is as follows:

$\mathrm{\&lambda;}=\mathrm{\&sigma;}\sqrt{2\log N}/\sqrt{N}---\left(7\right)$

This is an empirical equation, be not for all application, the threshold value of choosing according to this formula all can have good effect, generally, for different application, need us to reselect the suitable threshold value that should use, but top Donoho empirical equation can provide a basic reference for us.

A lot of scholars are by studies confirm that, generally, the method for hard-threshold can be introduced bigger variance, and the method for soft-threshold then can be introduced bigger deviation.Taken all factors into consideration the influence to variance and deviation, the scholar who has has also proposed the method that soft-threshold and hard-threshold combine.

4) wavelet reconstruction

The purpose of wavelet reconstruction is exactly to return time domain from the wavelet transformed domain conversion through wavelet coefficient after the threshold process and untreated scale coefficient.This reconstruct realizes that by the reconfigurable filter group it comprises a high pass and a low pass, and low pass filter and high pass filter coefficient are respectively ${\tilde{f}}_L=\left[\begin{array}{cc}\frac{\sqrt{2}}{2}& \frac{\sqrt{2}}{2}\end{array}\right]$ With ${\tilde{f}}_H=\left[\begin{array}{cc}\frac{\sqrt{2}}{2}& -\frac{\sqrt{2}}{2}\end{array}\right].$ Channel impulse response sample value sequence after the reconstruct Can be expressed as:

${\hat{h}}_{2k,l}=h_{k,l}^L{\tilde{f}}_L\left(1\right)+{\stackrel{\&OverBar;}{h}}_{k,l}^H{\tilde{f}}_H\left(1\right)$ $0\&le;k<\frac{M}{2}---\left(8\right)$

${\hat{h}}_{2k+1,l}=h_{k,l}^L{\tilde{f}}_L\left(0\right)+{\stackrel{\&OverBar;}{h}}_{k,l}^H{\tilde{f}}_H\left(0\right)$ $0\&le;k<\frac{M}{2}---\left(9\right)$

Be to the original channel impulse response sample value list entries h that has noise and interference _{K, l}Carry out noise and disturb removal output sequence afterwards.

With

Represent even number and progression sampling point output sequence value respectively.

The wavelet reconstruction process as shown in Figure 6.

5) Fourier transform

To and disturb the channel impulse response sequence transformation of removing to arrive frequency domain through noise:

${\hat{H}}_{k,l}=\underset{n=0}{\overset{N-1}{\mathrm{\&Sigma;}}}{\hat{h}}_{n,l}{e}^{-j2\mathrm{\&pi;nk}/M}$ 0≤k≤M-1 (10)

Wherein, M is the number of the pilot sub-carrier in an OFDM (OFDM) symbol, and the output sequence behind the process Fourier transform can be used for the interpolation on the frequency direction.

Embodiment

Actual DRM (digital radio globalization) broadcast channel is the frequency selectivity that existing multipath causes, two fading channels of the time selectivity that Doppler expands or frequency displacement causes are arranged again.Wherein multipath transmisstion is mainly because the ionospheric reflection of differing heights causes that the expansion of the maximum delay of channel can reach several milliseconds, and Doppler's expansion and frequency displacement cause mainly due to the spectral characteristic of ionospheric reflection and receiver mobile.With the mid latitudes is example, and the maximum of time delay expansion can reach 6ms, and Doppler's expansion then can be up to 5Hz.Generally, the representative value that time delay expansion and Doppler expand is 2ms and 1Hz, and this is the parameter value of our channel model 4 used just.

In line with the principle of proceeding from the reality, we have investigated the situation of channel model 3 and channel model 4 in DRM (digital radio globalization) standard, wherein channel model 3 is aimed at USConsortium (Consortium of United States) model of intermediate frequency and high frequency, and channel model 4 is aimed at standard CC IR (the international wireless electricity committee) model of high frequency.The concrete parameter of channel model 3 and channel model 4 provides in table 3 and table 4 respectively:

Compare with traditional linear interpolation method (on time orientation and frequency direction, being linear interpolation), the method that combined with wavelet transformed domain noise and interference are removed has better bit error rate characteristic than conventional method, this point can be come from our simulation result to channel model 3 and channel model 4 as can be seen, and better than the performance in conjunction with the hard-threshold method in conjunction with the performance of soft-threshold method.

Table 3: the parameter setting of channel model 3

	Path 1	Path 2	Path 3	Path 4
					Postpone	0	0.7 millisecond	1.5 millisecond	2.2 millisecond
Path gain	1	0.7	0.5	0.25
					Doppler frequency shift	0.1 hertz	0.2 hertz	0.5 hertz	1.0 hertz
Doppler's expansion	0.1 hertz	0.5 hertz	1.0 hertz	2.0 hertz

Table 4: the parameter setting of channel model 4

	Path 1	Path 2
			Postpone	0	2 milliseconds
Path gain	1	1
			Doppler frequency shift	0	0
Doppler's expansion	1 hertz	1 hertz

In our DRM (digital radio globalization) simulator, we adopt on the basis of same linear interpolation at time orientation, have investigated and have adopted the conventional linear interpolation method on the frequency direction and adopt hard-threshold and soft-threshold wavelet noise and the performance of disturbing the removal method.Wherein, the hard-threshold method as shown in Equation (4), the soft-threshold method is that formula (5) is carried out adjusted formula, is provided by formula (11).Concrete simulation parameter is as shown in table 5:

${\stackrel{\&OverBar;}{h}}_{k,l}^H=\left\{\begin{array}{cc}0& \left|h_{k,.l}^H\right|\&le;\mathrm{\&lambda;}\\ h_{k,l}^H-\mathrm{\&lambda;}& \mathrm{\&lambda;}\&le;h_{k,l}^H<2\mathrm{\&lambda;}\\ h_{k,l}^H+\mathrm{\&lambda;}& -2\mathrm{\&lambda;}\&le;h_{k,l}^H<-\mathrm{\&lambda;}\end{array}\right.---\left(11\right)$

Table 5: simulation parameter setting

Transmission mode	B
		Bandwidth	10K
Mapping mode	64QAM
		Code check	0.6
Channel model	Channel			3 and channel 4

Simulation result to channel model 3 and channel model 4 is distinguished as shown in Figure 7 and Figure 8.From simulation result we as can be seen, better in conjunction with hard-threshold wavelet field noise and method that disturb to remove, and in conjunction with soft-threshold wavelet field noise with disturb the method for removing more much better than method in conjunction with hard-threshold than the method for conventional linear interpolation.

For channel model 3, compare with the conventional linear interpolation method, in conjunction with the method for hard-threshold the gain of 0.3-0.4dB is arranged probably, the gain of 0.6-0.7dB is probably arranged in conjunction with the method for soft-threshold; For channel model 4, compare with the conventional linear interpolation method, in conjunction with the method for hard-threshold the gain of 0.3dB is arranged probably, the gain of 0.5-0.6dB is probably arranged in conjunction with the method for soft-threshold.

Claims (8)

1. the OFDM channel estimation methods of an eliminating noise combined with wavelet transformed domain comprises step:

A) information on the extraction pilot sub-carrier, the channel frequency response at calculating pilot sub-carrier place;

B) according to the channel frequency response at the pilot frequency locations place that calculates, channel frequency response is carried out interpolation at time orientation;

C) channel frequency response to the time orientation interpolation carries out noise and disturbs removal;

D) channel frequency response of having removed noise and interference is carried out the interpolation of frequency direction in the channel direction.

2. method according to claim 1 is characterized in that described step c) comprises:

The channel frequency response that the time orientation interpolation is obtained carries out inverse Fourier transform;

Carry out wavelet decomposition;

Carry out threshold process;

Carry out wavelet reconstruction;

Carry out Fourier transform, frequency domain is returned in the data conversion after handling.

3. method according to claim 2 is characterized in that described wavelet decomposition adopts the Harr wavelet decomposition.

4. method according to claim 3, it is characterized in that described Harr wavelet decomposition comprises high pass filter and low pass filter, wherein, output after list entries and the low pass filter convolution b referred to as scale coefficient, and the output after list entries and the high pass filter convolution is known as wavelet coefficient.

5. method according to claim 2 is characterized in that described threshold process comprises one of hard-threshold method, soft-threshold method and non-negative threshold value method.

6. method according to claim 5 is characterized in that described threshold value chooses by following formula:

$\mathrm{\&lambda;}=\mathrm{\&sigma;}\sqrt{2\log N}/\sqrt{N}$

Wherein, λ represents threshold value, and λ 〉=0, and σ represents the noise size, and N is the data length that needs processing.

7. method according to claim 6 is characterized in that: choose suitable threshold, will be lower than the whole set of wavelet coefficient zero of certain threshold value, what bigger wavelet coefficient comprised is useful channel information.

8. method according to claim 2 is characterized in that described wavelet reconstruction will be through the wavelet coefficient after the threshold process and untreated scale coefficient threshold when wavelet domain transform is returned.

Images