The Study Of A TEQ Algorithm And Structure For DMT-ADSL Systems

Modern communication technology is being developed towards broad band , integration , digitization and individuation. Because of low cost and high speed which have been widely concerned, Asymmetrical Digital Subscriber Loop (ADSL ) modulation technology is the best way to carry on broadband transmission utilizing the existing access network (twisted wire-pair) at present. ADSL system uses time domain equalization and cyclic prefix (CP) to eliminate inter-symbol interference (ISI) and inter-channel interference (ICI). This paper emphatically studies a effective time domain equalization algorithm based on ADSL-DMT system,It brings up the dual-path time domain equalizer structure and then draw the conclusion by computer emulation. 1. Summary of time domain equalization in DMT system The experiment has proved: the impulse response of many channels are very similar to the impulse response of zero-poles model. we can master the channel characteristic through the examining zero-poles model. The tested zero-poles model of the channel is supposed to: A (Z ) , P (Z ) are all FIR filter forms . If a FIR time domain equalizer is connected behind the receiver and ahead the FFT, then the impulse response of " shorten channel" after equalization can be shorten to length of A (Z ). If all parameters of A (Z ) are get , the equivalent response of channel which contains H (Z) and A (Z) will approximate to be B (Z ) ,then it is a FIR filter with very short impulse response. Set up h(n) as the impulse response of dispersed time channel, n =1,2,3 ……,time domain equalization is to design a FIR filter w(n) which is used to equalize the impulse response of the channel as: Suppose the length of the channel's impulse response which have been time domain equalized is (v+1), and the sending information code is N.Then the length of non-zero values after b (n) convoluts x (n) is (N+V) .It will cause the overlap of V between the last symble and the next symbol,then the ISI appears. It is obvious that the interference can not be eliminated by bringing in time domain equalizer only. In order to guarantee every subchannel one is separated and dispel ISI , DMT system duplicates the data of certain length of one yard of tails of information before yard, this part of information duplicated is called the cyclic prefix (CP). The goal of time domain equalization in DMT system is to compress the channel impulse response within the length of CP and control the length of CP. Time domain equalization algorithm can be divided into three kinds of algorithm, namely Minimum Mean Squared Error (MMSE), Maximum Shortening SNR algorithm (MSSNR) and Maximum Geometry SNR algorithm (MGSNR). 2.A LMS time domain equalization algorithm based on MMSE Among the lots of time domain equalization algorithms, the application of MMSE algorithm is more general, but it's complexity is high, involved Matrix inversion autocorrelation,corosscorrelation,eigenvalue decomposition and so on. Unit's energy restraining needs search optimum value correctly. So Falconer & Magee combines MMSE, has proposed a algorithm (LMS) based on minimum mean square. Chow , Cioff and Bingham have put forward the method to refresh SIR , TEQ further: frequency domain LMS and frequency domaint decomposition . In order to guarantee the length of SIR and TEQ meets the demands after refreshing, we will transform SIR to time domain and add the window. This paper has carried on the computer emulation of method on three kinds of loop circuit the emulation result shows : (1)In DMT system, the method combines time domain equalization with cyclic prefix is used to eliminate inter-symbol interference .It makes system performance decline in a certain range while avoiding too many time delay tap. It's a effective method dispells ISI in ADSL-DMT system at present. (2)The energy ratio after windowing:energy ratio between the (v+1) samples and all energies is greater,the system performance is better.It is generally greater than 95%. (3)In this paper, we simplify adding windows after refreshing w to set up initial value. The simulated result is similar to the result of traditional algorithm. 3.Dual-path time domain equalizer based on ADSL-DMT system 3.1 Dual-path time domain equalizer based on ADSL-DMT system The structure of DMT receiver's dual-path TEQ is as Fig.1. For each audio, we choose one of the two paths as output according the presuppositional rule. The standard of choosing path is that for each audio there is higher sub channel SNR afterFEQ, that is to say for each audio the output from FEQ is x i = ??? xx ii21 ifSN其R它i1>SNRi2 Fig. 1 dual-path TEQ structure One choice is that each TEQ optimizes one different part of bandwidth and subchannels stretch over the whole bandwidth. But, this part is very difficult to optimize most. So we choice a little different approximate as follows: (1) Path1 is an ordinary TEQ path , it optimizes the whole bandwidth. (2) Develop another TEQ, it optimizs the subchannel when preseting the window in a frequency in route 2. Subchannel with owns higher SNR usually have larger space to improve bps by the equalization of second path. So these subchannels should be put into window. One simple subchannel choosing method is to let a window slip through all the subchannels, the initial audio index i is determined by the formula (3-1 ). ∑=+?== kiWL1kin,i2i start argmaxiSx,iHi/S (3-1) Where WL is the length of window, S x,i,S n,iand H iare the power of transmit signal, the power of channel noise and the frequency respond of the ith audio separately. Any algorithm can be used to train the first TEQ,but there are only a few TEQ design methods have some control to frequency of response of TEQ frequency. They can design the second TEQ. Among them, the Maximum Bit Rate (MBR) and minimum ISI are more suitable for choosing subchannel. The basic thought of MBR and minimize ISI is dividing the output of TEQh(n)*w(n) into the useful signal part and ISI part by adding window. TEQ is exported y(n)=h(n)*w(n)*x(n)+w(n)*v(n) = h signal(n)*x(n)+ h ISI(n)*x(n)+w(n)*v(n) (3-2) For each audio, SNR is defined and can be written n,i2x,iiISI2ix,isignal2iiHSWSSNR = H+S (3-3) where, His ignal,H iISI,W i are the ith DFT sample of h signal, ISIh ,w separately. Write it in matrix form as follows: n,iH2x,iiH2ix,iH2iiqDHwSqFHwSqGHwSSNR= + (3-4) where G=diag[g(0)g(1)…g(N-1)] T ,D=I-G H = ?????? h(Nhh(( M10 ?)) 1)h(hNh((?M0?)1)2)LLLO hhh(((???( ((NLLMww???L12w))))))?????? [ ]j2/j2i(N-1)/NTq i = 1eπiNLeπ(3-5) where, q iH FHwis the ith T coefficient of FFT (N-point) of w. MBR carries out the function of maximum bit allocation with non-linear optimization =∑+iib DMT log2(1SNΓR) (3-6) If the ISI number can be forced to 0 ,in (3-3) the SNR of subchannels will not depend on the the setting of equalization. This is the maximum SNR can be fulfilled. The bound of matched filter is SNR i = Sx,iHi2/Sn,i.3.2 Computer's emulation and analyzing This paper has carried out computer simulation for dual-path TEQ, the simulated curves are shown in Fig. 2 , Fig. 3 and Fig. 4, the analysis are as follows: (1)Channel noise is simulated as -140dBm AWGN distributed on whole bandwidth adds NEXT noise, the code gains is 4. 2dB, system clearance is 6dB, the power of the input signal is allocated in the subchannels which is 23dBm, FFT size is N =512. (2)Bit allocation has the same trend as SNR. It can be found out from Fig. 3 that some audio with higher SNR in Fig. 2, especially those audio with peak value , the second TEQ behaves better in bit allocation figure. There are different improvements in everywhere. (3)Dual-path TEQ increases the capacity of channel from 7. 2566 to 7. 3375Mbps which is up to 1. 1%, and bit allocation averagely from 8. 425 to 8. 527bit/subcarrier which is improved by 1. 214%. (4)If the designed technique is feasible, the dual-path TEQ reaches a higher bit rate than the single -path TEQ.