DESCRIPTION

Method and apparatus for cyclic reconstruction for frequency equalizer

Technical Field

The invention regards to a method for cyclic reconstruction for frequency equalizer according to pre-characterizing part of claim 1, and to an apparatus, especially frequency equalizer.

In a communication system, when signals come to a receiver, they have been distorted by noise and frequency selective fading. The noise is almost white and Gaussian distributed which can be coped with channel code. The frequency selective fading is caused by multi-paths. To combat the frequency selective fading there are two methods used widely. One is using a time domain equalizer which works in time domain to counteract the distortion direct which means finishing the channel estimation and equalization together. Another method is using of a frequency domain equalizer, which works in frequency domain to cancel the distortion by known channel- response coming from a channel estimation block.

For classical OFDM system, such as DVB-T, there is offered a cyclic prefix (CP) in every frame. Thus, the received data corresponds to a natural circle convolution of sent data with channel-response. It is very easy for the frequency equalizer to use a division to cancel the distortion in the received data .

For TDS-OFDM system, such as the terrestrial broadcasting standard of Chinese digital television, there is offered a

pseudo random (PN) sequence as guardian interval between every two frames. The received data are the linear convolution of sent data with channel-response. Since the natural circle convolution effect of received data part does not exist any more, cyclic reconstruction should be used to transform the linear convolution to circle convolution.

However, cyclic reconstruction introduces additional noise into the received data. In this method a skill of dynamic channel-length detection approach is used to reduce the additional noise as more as possible.

In a dynamic channel, the magnitude of some multi-paths will change at times, which will cause that themselves sometimes be detected by frame-synchronization detection block and sometimes not. If this multi-path is the first one, which can be detected, it will make the frame-synchronization shift and not occur in the fixed position. This method will cope with this trouble to keep the frequency equalizer work smoothly in dynamic channel.

Background Art

DVBT is a classical OFDM system, which has the frame structure as shown in Fig. 14.

Using x(n) to denote the sent data, y(n) to denote the received data, h(ri) to denote the channel-response which comes from the channel estimation block, because there is cyclic prefix CP in the front of every frame which is the copy of the tail of the current frame's data part, then y{ή) is natural circle convolution of x(n) with h(ή) . If y(n) and

h(n) are transformed from time domain to frequency domain by a FFT block (FFT: Fast Fourier Transformation) to obtain Y(k) and H{k) , according to the character of FFT transformation,

Y(k) can be written by equation

r(k)= ffi(y(n)) = ffi(x(n)®h(n)) =fβ(x(n)).fft(h(n)) =X{k)•H{k) ,

where "®" means a circle convolution, "•" means a multiplication point by point and d(ri) means d{ή) being a vector dimension of which is n. Obviously X{k) can be gotten after dividing Y(k) by H(k) . Here X(k) is the frequency domain description of sent data x(n) , supposed there is no noise in the received data y(n) . The distortion of channel- response in the received data y(n) is cancelled. In one sentence, the frequency equalizer is just based on the received data's circle convolution character.

In a traditional communication system, PN part is popular in frame structure which is widely useful to detect carrier synchronization, timing synchronization, frame synchronization, and channel estimation. In an OFDM system, if a PN part is added into the frame, it is called TDS-OFDM, as shown in Fig. 15. Frame starts with PN part followed by CP and data. Again CP corresponds to tail of data part.

This structure will not damage the natural circle convolution character of received data, but with the PN part added into every frame, the useful information rate reduces.

Chinese standard chooses a different frame structure as shown

in Fig. 16. PN part is followed directly by data part. Since there is no CP in frame, the received data part is not the circle convolution of sent data with channel-response any longer. Cyclic reconstruction is needed to construct the received data's circle convolution character.

Technical Problem

It is an object of the invention to provide another method and apparatus for cyclic reconstruction for frequency equalizer .

Technical Solution

This object is solved by a method for cyclic reconstruction for frequency equalizer having features according to claim 1, and by an apparatus for cyclic reconstruction for frequency equalizer having features according to claim 10. Preferred aspects and embodiments are subject matter of dependent claims.

Especially, there is provided a method for cyclic reconstruction in a frequency equalizer in system with pseudo random sequence and without cyclic prefix, wherein data consisting of linear convolution of sent data with channel- response are received, and wherein received data are transformed to a circle convolution of sent data and of channel-response .

Further, there is provided an apparatus in a TDS-OFDM system, especially system according to terrestrial broadcasting standard of Chinese digital television, comprising means adapted for receiving data consisting of linear convolution of sent data with channel-response, said data offering a

pseudo random sequence as guardian interval between every two frames, a channel estimation block adapted for detecting a length of the channel-response by channel estimation, a channel-length detection block adapted for canceling a distortion of linear convolution remaining in the received data, an information compensation block adapted for compensating an information of circle convolution which leaks out of the received data, and a starting position decision block adapted for generating a starting position of cyclic reconstruction depending on a frame-synchronization.

Advantageous Effects

Especially there is preferred a method, comprising the steps of detecting a length of the channel-response by channel estimation, canceling a distortion of linear convolution which remains in the received data, compensating an information of circle convolution which leaks out of the received data, and generating a starting position of cyclic reconstruction. Especially, such method can be executed independent to any format of pseudo random sequence by executing these four main functions.

The channel-length detection can judge the channel-response every frame. Especially, there is detected an effective length of channel-response, which is not longer than a length of pseudo random sequence part. Using such valve, since cyclic reconstruction will introduce additional noise proportional to the channel-length, this dynamic channel- length detection approach can reduce such noise enhancement as more as possible. Distortion canceling can be executed depending on a current frame's local pseudo random sequence and channel-response. Information compensation can be executed depending on a next frame' s local pseudo random

sequence, channel-response and received pseudo random sequence part.

Starting point decision is executable depending on a frame- synchronization, wherein the frame-synchronization is executed by shifting in dynamic channel, and both the distortion canceling and information compensation are based on current frame's location.

All of multi-paths in the channel-response can be considered as postfix multi-path to get good results.

Especially, there is provided such an apparatus comprising a distortion canceling block adapted for distortion canceling and information compensation basing on current frame' s location and adapted for executing a frame-synchronization by shifting in dynamic channel.

Such apparatus can be adapted for executing such a preferred method.

Description of Drawings

An embodiment will be disclosed in more details with respect to enclosed drawing. There are shown in:

Fig. 1 a composition of received data in PN420 mode of Chinese standard,

Fig. 2 a received data part and an ideal result of cyclic reconstruction,

Fig. 3 a block diagram of equation 1,

Fig. 4 a block diagram of equation 2,

Fig. 5 a block diagram of equation 3,

Fig. 6 a block diagram of equation 4,

Fig. 7 a frame-synchronization shift behind in dynamic channel,

Fig. 8 a block diagram of equation 5,

Fig. 9 a block diagram of equation 6,

Fig. 10 a frame-synchronization shift front in dynamic channel,

Fig. 11 a block diagram of equation 7,

Fig. 12 position of cyclic reconstruction,

Fig. 13 a structure of cyclic reconstruction arrangement,

Fig. 14 a classical OFDM system which has the frame structure of DVBT,

Fig. 15 TDS-OFDM system, having a PN part which is added into its frame, and

Fig. 16 different frame structure of Chinese standard.

Mode for Invention

According to preferred embodiment there is provided a method for cyclic reconstruction for frequency equalizer and an

equalizer executing such method. Method bases on transformation of a linear convolution to a circle convolution by cyclic reconstruction. For convenience to analysis, the pn420 mode of Chinese standard, which has 420 symbols of PN part and 3780 symbols of data part, is used to explain the method. However, method can be executed in systems according to other standards, too.

In the Figures within a frame 1, noise 2 is shown by dotted signs, linear convolution of send data parts with channel- response is shown by cross hatched signs, linear convolution of send PN parts with channel-response is shown by hatched signs, clean PN with a number n of symbols is shown by empty or white colored symbols and text "PN (n)", and clean data with a number n of symbols is shown by empty or white colored symbols and text "data(n)".

Fig. 1 shows a composition of received signals, which can be written as y(n). In PN420 mode of Chinese standard, the length of every frame is 4200 symbols, and the received symbols are linear convolution of sent symbols with channel- response. All the multi-paths in the channel-response are considered as postfix.

Fig. 1 discloses a diagram showing two frames, a first and current frame 1, y(n) and a following second and next frame 1°. In the Chinese standard each of the frames 1, 1° consist of 4200 symbols. The first section of each frame 1, 1° consists out of a header or frame head. When sending out such frames 1, 1° the first section will be without noise, too.

Especially, the head will consist out of a clean PN with 420 symbols length or duration. In addition, there might be a tail of a preceding frame consisting of linear convolution of sent data part. After receiving such frames the received data

parts consist out of a linear convolution of the sent data part with channel-response 3. Further, the head consist of a linear convolution of sent PN part with channel-response 4. In addition the symbols of the head and the symbols of the data part are covered with noise, too.

According to a preferred embodiment the head consists of a linear convolution of sent PN part with channel-response consisting out of a number symbols or length Ni, N ₂ , wherein this number of symbols is smaller than a theoretically usable maximum number of 420 symbols. According to a preferred embodiment the head consists of a linear convolution of sent PN part with channel-response consisting out of a number symbols or length Ni, N ₂ , wherein this number of symbols is smaller than a theoretically usable maximum number of 420 symbols. In addition, Ni means the length of the frame l's channel-response, it starts at the position of frame synchronization and ends at the last multi-path which is larger than the threshold of channel-response, so that the length of channel-response is kept as its minimum possible value which can maintain it's performance and minimize the introduced noise during reconstruction, so as N ₂ . (note: the not linear increasing and decreasing shape is just used to show there is an multi-path in channel-response, this shape does not mean anything more.)

In the following, there are used some abbreviations to explain execution of the preferred method. Signs are defined as below: - pn_c(420Y ={pn _Q _c ^r ,pn _] _c ^r ,---,pn _4]9 _c ^r } denotes the current received frame' s PN part according to PN420 mode of Chinese standard;

- pn_c{A20) ^s = {pn ₀ _c ^s ,pn _x _c ^s , - -,pn _4l9 _c ^s } denotes the sent PN part

corresponding to pn_c(42Q) ^r ;

- data_c(3780Y={data ₀ _c ^r ,data _λ _c ^r ,---,data _il99 _c ^r } denotes the current received frame's data part;

- data_c(37%θy ={data _o _c ^r ,data _} _c ^r ,---,data ₃₇₉₉ _c ^r } denotes the result of data _c(3780) ^r after cyclic reconstruction;

- DATA _c(3780) ^r = {DATA ₀ _c ^r ,DATA _λ _c ^r ,-• ^■ , DATA ₃₁₉₉ _c ^r } denotes the description of data_c(3780) ^r in frequency domain;

- data_c(37$0Y = {data ₀ _c ^s ,data _: _c ^s ,■ --,data ₃₁₉₉ _c ^s } denotes the sent data part corresponding to data _c(3780/ ; - pn_n(42θy={pn _o _n ^r ,pn _] _n ^r ,---,pn _4]9 _n ^r } denotes the next received frame' s PN part received as frame following the current received frame's PN part pn_c(420) ^r

- pn_n(420y ={pn _Q _n ^s , pn _λ _n ^s ,-- -,pn _m _n ^s } denotes the sent PN part corresponding to pn_n(420) ^r ; - data_n(37SO)" ={data ₀ _n ^r ,data _x _n ^r ,---,data ₃₇₉₉ _n ^r ) denotes the next received frame's data part;

- data_n(3780) ⁱ = {data ₀ _n ^* ,data _λ _n ^s ,-•■,data _il99 _n ^s } denotes the sent data part corresponding to data_n(37S0) ^r ;

- che_c(n _l ) = {che _Q _c,che^ _c,---,che _ri[ _ _] _c) denotes the channel estimation of the current received frame; and

- che_n(n ₂ ) = {che ₀ _n,che _] _n,---,che _n^ _ _] _ri) denotes the channel estimation of the next received frame.

In Fig. 2, the left one is the received data part 7, which includes three parts: noise 2, the linear convolution of sent data with channel-response 3 which remains in the received

data and the linear convolution of sent PN with channel- response 4* which folds into the received data. The right one is ideal result of cyclic reconstruction having no PN part 4* but noise and the linear convolution of sent data with channel-response (which include both 3 and 3*) after reconstruction.

Fig. 3 shows a first equation 1 in symbolic form. data_c(37SOY can be gotten by the steps shown by Equation-1. From the received data part 7 there is subtracted the linear convolution of sent PN with channel-response 4* and added the linear convolution of sent data with channel-response 3* to get data_φ7S0) ^r 8.

Obviously, the cyclic reconstruction includes two steps, firstly removing the linear convolution of sent PN with channel-response 4*, from the head of data_c(37%0) ^r , which can be called "distortion cancelling"; secondly transferring the linear convolution of sent data with channel-response 3*, B^w from the head of pn_n(420) ^r to the head of data_c{37%Q) ^r , which can be named "information compensation".

(Note: both and is the linear convolution of sent data with channel-respnse, remains in the part of received data, and §iS&& _^ folds into the received next frame's PN part. )

For distortion cancelling, there is used the tail of linear convolution of PN with channel-response 4* (note: which folds into the received data part, is the tail of linear convolution of pn_c(420) ^s with che_c , according to the step of

removing it is named as "distortion cancelling", it can be called "redundant distortion") . Because both the channel estimation che_c and pn_c(420Y are known, this is easy to calculate .

For second step which is named information compensation, there is used the tail of linear convolution of sent data part with PN 3*, (note: .^-Ss--. which folds into the received next frame's PN part, is the tail of linear convolution of data_c(3780) ⁱ with the channel estimation che_c , according to the step of compensating it is named as "information compensation", it can be called "leakage information").

Because data_c(37$0) ^s is unknown, it cannot be calculated direct. But it is included in the received PN part pn_n(420y , so its indirect calculation is reasonable as

Equation-2. Using symbolic signs in Fig. 4 shows equation 2

From the equation upper, the received PN part pn_n(420) ^r is known. The head, i.e. the first 420 symbols, of the linear convolution of the sent PN part pn_c(420) ^s with the channel estimation che_n is easy to be calculated. So after subtraction, 3*, -S88&J&.. (leakage information) will be gotten.

Unfortunately, the noise in the received PN part pn_n(420) ^r will be introduced into the result at the same time. And obviously if the length of channel-response is longer than 420 symbols, will fold into the current received frame's data part data_n(3780) ^r . Because the sent data part data_n(3780Y corresponding to the current received frame ' s data part data_n(37S0) ^r is not known, the sent data part

can not be calculated by the same way, which means that cyclic reconstruction can not deal with the channel-response longer than the length of pseudo random sequence PN.

Totally, data_c(37S0) ^r being the result of data_c(3780) ^r after cyclic reconstruction can be calculated by Equation-3. Fig. 5 shows equation 3 in symbolic signs.

Thereafter, the additional noise is reduced by channel-length detection. From the analysis above, additional noise will be introduced into data_c(3780) ^r being the result of data_c(37%0) ^r after cyclic reconstruction in the processing of information compensation. If consider the length of channel-response as the maximum value, the power of noise in the first 420 symbols will statistical be 2 times than the other symbols in data_c(378Q) ^r after cyclic reconstruction.

After FFT block, noise will be scattered into all the symbols in DATA_c(37S0) ^r , so the signal to noise relation (SNR) of the symbols in DATA_c(3780) ^r will be worsened by 0.46 dB

(10*logl0( (3780+420) /3780) = 0.46 dB) . The loss is a little bit large. But for most cases, the effective length of channel-response is much less than the length of PN. For example, Chinese standard defines the symbol rate as 7.56 MHz, and the longest channel model is 35 μs . Even considering the multi-paths will span in channel estimation, the effective length of channel-response is less than 300 symbols. So in information compensation block, if there are only used the first 300 symbols in the next received frame' s PN part pn_n(420) ^r to attend reconstruction, the SNR loss will be 0.33 dB ( 10*logl0 ( (3780+330) /3780) = 0.33 dB) , which

is 0.13 dB better 0.46 dB. In other channel models, the benefit is much higher.

So for cyclic reconstruction, firstly there should be analysed the channel-response which comes from channel estimation block, set a valve to keep the paths of channel- response which are less than the valve to be zero, because these small paths are generated by system noise or algorithm noise with high possibility. Without these small paths the estimated channel-response will be more accuracy. Then, the length of channel-response should be counted from the first point of channel-response to the last point larger than the valve. The length denotes as N, and the length should be smaller than the length of the head, i.e. N < 420.

Secondly, the cyclic reconstruction is finished as shown in Equation-4 by Fig. 6.

Furthermore according to a preferred embodiment, because the length of channel-response is judged every frame, the length of adjacent frame's channel-response may be different. According to the rule of using the latest channel-response, the length of channel-response should be used when calculating linear convolution of sent PN with channel- response 4*, fcfc»»_ (redundant distortion) which is according to the current received frame, and when calculating the linear convolution of sent data with channel-response 3*, leakage information) the length of channel-response should be used which is according to the next received frame. This difference can be seen in the upper equation 4 in Fig. 6.

By this improvement, the SNR loss is not decided by the length of PN but proportional to the length of the next

frame's channel-response, which is much shorter than the length of PN. So the threshold of system will not reduce much.

According to a further aspect, the problem of frame- synchronization shift has to be handled. In dynamic channel, the magnitude of some multi-paths will change at times, which will cause themselves sometimes be detected by frame- synchronization detection block and sometimes not. If this multi-path is the first one, which can be detected out, it will make the frame-synchronization shift and not occur in the fixed position.

When comparing Fig. 7 with Fig. 1 it can be seen that there might be a virtual frame synchronization of next frame starting with a distance of example 10 symbols behind current frame. Thus, first lower amplitude maximum followed by the residual maxima will fall within the duration of the ten symbols, which are a virtual frame sync of the next frame. Thus, only second amplitude maximum with the duration of length N ₂ will reach into the head section of the next frame. Thus, Fig.7 shows result of equation-4, which is given in Fig. 6.

In Fig. 7 shows a frame-synchronization shift behind in dynamic channel. When the first multi-path, which can be detected in the current frame, becomes too weak to be detected in the next frame, the frame-synchronization generated by frame detection block will shift behind by 10 symbols that mean the loss multi-path spans with 10 symbols in width.

In this case, if the information compensation works based on the frame-synchronization of next frame as Equation-5 being

shown in Fig. 8 neither the position nor magnitude of the leakage information &S?δ!ftir> compensated into the current received frame's data part data_c(37S0) ^r is what we want. This would make the restored part in data_c(3780) ^r wrong, and after FFT block the error is scattered into the whole result of data_c(3780)' after cyclic reconstruction DATA_c(37S0) ^r , what makes the current received frame's data part totally crash.

Thus, Fig. 8 highlights result giving a shorter real compensation of the tail of the linear convolution of sent data part with channel-response compared with ideal compensation.

So, the right way is that both of the two blocks should work based on the frame-synchronization of current frame. In details, the distortion cancelling should work based on the frame-synchronization of current frame and the information compensation should work based on the virtual frame- synchronization which has fixed distance, especially the fixed length of a frame, from the current one. The method is described as Equation-6 and shown in Fig. 9.

The figure and equation when the frame-synchronization shift front is shown in Fig. 10, which are similar with the shifting behind. Fig. 10 shows the frame-synchronization shift front in dynamic channel. When comparing Fig. 10 with Fig. 1 it can be seen that current frame will have only 4190 symbols, i.e. 10 symbols less than current frame of Fig. 1. In consequence amplitude in starting section of next frame will have three maxima, two low amplitude maximum values followed by one higher amplitude maximum value within the second duration or length N ₂ . In such situation the virtual

frame sync of the next frame will start ten symbols after the frame sync signal of the next frame. Thus, there is a rising multi-path within the first ten symbols followed by the same amplitude structure like during first duration Ni.

Equation-7 being shown in Fig. 11 makes possible to draw a conclusion that the method of cyclic reconstruction mentioned in this description can not only transform the received data part from linear convolution to circle convolution with less noise introduced, but also combat the frame-synchronization shift in dynamic channel. It will help the frequency that the equalizer has good and stable performance in all kinds of channels, which can be used widely in communication system.

According to a further aspect there is described the position of cyclic reconstruction in system and its structure. The position of cyclic reconstruction in the system is shown in Fig. 12, which is behind channel estimation block and before FFT&FEQ. In Fig. 12 data are inputted into a channel estimation block 30. Channel estimation block 30 outputs data and a channel estimation che. Data and channel estimation signals are outputted from channel estimation into a block for cyclic reconstruction 31. After cyclic reconstruction data and channel estimation che are outputted into a block FFT 32 executing a fast Fourier transformation. This block FFT outputs data and channel estimation che to a FEQ block 33, which executes a forward error correction FEQ. This FEQ block 33 outputs data and a channel estimation signal.

The structure of cyclic reconstruction is shown as Fig.13.

The function of the sub-blocks in an arrangement according to the block diagram is described as above. Fig. 13 shows a channel-length detection block 34. Inputted into the channel- length detection block 34 are data and a channel estimation

signal che. After channel-length detection data and channel estimation che are outputted out of the channel-length detection block 34 and are inputted into a distortion- cancelling block 35. Further there is inputted a start signal generated in a starting position decision block 36 into the distortion-cancelling block 35. Data and channel estimation che outputted from the distortion cancelling block 35 and in addition a start signal from the starting position decision block 36 are outputted into an information compensation block 37 outputting compensation result by outputting data and a channel estimation signal. Starting position decision block 36 gets frame synchronization signal frame sync as input signal.