hi, i am fresh in the field of ADSL/ADSL2 modems. our system has 2-stage equalization - Time-domain equalizer for reducing ISI to within the cyclic prefix, and Frequency-domain equalizer which is a 1-tap equalizer to undo the channel. for training the TEQ, a signal called REVERB is used, a psuedo-random signal which is identical from one symbol to the next. for training the FEQ and estimating the SNR, a wideband psuedo-random signal called MEDLEY is used, which is different from one symbol to the next (for at least 511 symbols, after which the sequence of symbols is repeated) my question is - why is it necessary to use MEDLEY for calculating SNR? why cant we use the REVERB signal itself? Regards, Manish. This message was sent using the Comp.DSP web interface on www.DSPRelated.com

# estimation of SNR using MEDLEY

Started by ●May 19, 2005

Reply by ●May 24, 20052005-05-24

manish schrieb: (snip)> my question is - why is it necessary to use MEDLEY for calculating SNR? > why cant we use the REVERB signal itself?Did you ever find an answer to this question? Guenter

Reply by ●May 26, 20052005-05-26

no i havent found an answer. would really appreciate some help on this. Regards, Manish. This message was sent using the Comp.DSP web interface on www.DSPRelated.com

Reply by ●June 10, 20052005-06-10

manish schrieb:> no i havent found an answer. would really appreciate some help on this.I am a bit slow here, I was doing some searching and found this explanation of the MEDLEY: "Because this polynomial (the one for MEDLEY)* produces a pseudorandom binary sequence with period 511, but 512 bits are needed for a symbol, the postion of the pseudorandom bits and the symbol boundary will slide from symbol to symbol, thus rendering the appearance of uncorrelated successively transmitted MEDLEY symbols. Note this differs from REVERBERATION, which repeats the same symbol." - John Cioffi & Jacky Chow, Amati, "Recommended Training Sequence for SNR Computation with DMT ADSL", T1E1.4/93-087 * my note So the explanation seems to be that for the SNR calculation the signal needs to be uncorrelated from symbol to symbol? Maybe you have some insight into that and can give an explanation why? Guenter

Reply by ●August 5, 20052005-08-05

thanks Guenter. you are right. i also did some reading, and the answer is in basic information theory......the bitloading equation b=log2(1+S/N) makes the following assumptions - 1. Noise is AWGN. 2. Signal has the same probability distribution as noise. 3. signal and noise are statistically independent. thus, to measure SNR accurately, we need a signal which is as random as possible, and thats why MEDLEY is used. Regards, Munish.>"Because this polynomial (the one for MEDLEY)* produces a pseudorandom >binary sequence with period 511, but 512 bits are needed for a symbol, >the postion of the pseudorandom bits and the symbol boundary will slide >from symbol to symbol, thus rendering the appearance of uncorrelated >successively transmitted MEDLEY symbols. Note this differs from >REVERBERATION, which repeats the same symbol." > >- John Cioffi & Jacky Chow, Amati, "Recommended Training Sequence for >SNR Computation with DMT ADSL", T1E1.4/93-087 > >* my note > >So the explanation seems to be that for the SNR calculation the signal >needs to be uncorrelated from symbol to symbol? > >Maybe you have some insight into that and can give an explanation why? > > >Guenter > >This message was sent using the Comp.DSP web interface on www.DSPRelated.com