Forums

ADSL Demodulation

Started by DougB May 18, 2012
Hello,

In every book and article aobut ADSL processing the A/D and D/A sample
rates are given as 2.208 MHz.  For a transmitter implementation at this
sample rate you would get the conjugate symmetric image spewing out of your
D/A since one may use up to bin 255 and there is absolutely no
oversampling, and no analog filter in the world can cut-off that fast. 

Likewise for the receiver, any signal, interference, and noise above 1.104
MHz will fold/alias into the digitized signal because no anti-aliasing
filter in the world will cut off fast enough to prevent this.

So my question is:  Is this really how ADSL modems are implemented?  One
solution would be to double the DFT size, thus oversampling the signal,
which allows enough bandwidth for the analog filters to cut off; however I
never see any reference to such techniques being used.

The only other explaination is that you just take the loss and
performance.

So I was wondering if anyone can enlighten me :).

Thanks
-Doug


On Fri, 18 May 2012 08:58:05 -0500, DougB wrote:

> Hello, > > In every book and article aobut ADSL processing the A/D and D/A sample > rates are given as 2.208 MHz. For a transmitter implementation at this > sample rate you would get the conjugate symmetric image spewing out of > your D/A since one may use up to bin 255 and there is absolutely no > oversampling, and no analog filter in the world can cut-off that fast. > > Likewise for the receiver, any signal, interference, and noise above > 1.104 MHz will fold/alias into the digitized signal because no > anti-aliasing filter in the world will cut off fast enough to prevent > this. > > So my question is: Is this really how ADSL modems are implemented? One > solution would be to double the DFT size, thus oversampling the signal, > which allows enough bandwidth for the analog filters to cut off; however > I never see any reference to such techniques being used. > > The only other explaination is that you just take the loss and > performance. > > So I was wondering if anyone can enlighten me :).
If the transmitter and receiver ADCs are synchronized, then the signal folding will not be nearly as much of a problem (this is obvious in the time domain; if you grind through all the frequency-domain calculations while dotting all the 'i's and crossing all the 't's correctly you'll find that the signal aliases all add constructively). Interference and noise will be an issue, but an analog filter _can_ at least cut the amount of interference and noise, even if it can't cut all possible interference and noise. It sounds like you're stuck in textbook communications engineering, which goes to great lengths to do optimization on signal/noise ratio, but only mentions optimizing performance/cost in passing. I rather suspect that the theory behind the ADSL coding is that, yes, they aren't squeezing every last bit of capacity out of the physical channel, but that no, they aren't still stuck in the lab decades later, trying to come up with a way to implement The Best Modulation Scheme Ever for a price that consumers are willing to pay. -- My liberal friends think I'm a conservative kook. My conservative friends think I'm a liberal kook. Why am I not happy that they have found common ground? Tim Wescott, Communications, Control, Circuits & Software http://www.wescottdesign.com
On Fri, 18 May 2012 08:58:05 -0500, DougB wrote:

> In every book and article aobut ADSL processing the A/D and D/A sample > rates are given as 2.208 MHz. [...] > So my question is: Is this really how ADSL modems are implemented?
I haven't worked on an ADSL modem, but my guess is the answer to your question is, almost certainly not. It a typical parallel tone (OFDM) modem implementation, the sample rates are either 2X or 4X the sample rates given in the standards document. Steve
Tim Wescott <tim@seemywebsite.com> wrote:
> On Fri, 18 May 2012 08:58:05 -0500, DougB wrote:
>> In every book and article aobut ADSL processing the A/D and D/A sample >> rates are given as 2.208 MHz. For a transmitter implementation at this >> sample rate you would get the conjugate symmetric image spewing out of >> your D/A since one may use up to bin 255 and there is absolutely no >> oversampling, and no analog filter in the world can cut-off that fast.
(snip)
>> The only other explaination is that you just take the loss and >> performance.
>> So I was wondering if anyone can enlighten me :).
Well, for one the frequency range doesn't go all the way down to zero. http://en.wikipedia.org/wiki/Asymmetric_digital_subscriber_line#Operation If you filter out the PSTN and upstream part, the downstream is now somewhat less than 1104kHz wide, so the filter doesn't have to be so sharp.
> If the transmitter and receiver ADCs are synchronized, then the signal > folding will not be nearly as much of a problem (this is obvious in the > time domain; if you grind through all the frequency-domain calculations > while dotting all the 'i's and crossing all the 't's correctly you'll > find that the signal aliases all add constructively). Interference and > noise will be an issue, but an analog filter _can_ at least cut the > amount of interference and noise, even if it can't cut all possible > interference and noise.
In the case of digitizing an existing analog signal, such as in digital audio, one does have to worry about harmonics that might be up there, and alias down. While the amplitude might be small, new low frequency components would sound very strange. In digital cameras, aliasing can be visible, but maybe not quite as much of a problem as in digital audio. For one, the lens of cheaper cameras will act as a low pass filter. Better cameras uses an actual optical low-pass filter in front of the sensor. In this case, though, except for noise the signal is only that generated by the other end. If the noise level is low enough, one only has to consider the harmonics generated by the D/A converter, and those will be predictable. I believe that is related to what Tim said above. In addition, note how DSL uses the bandwidth. It is divided into many channles of 4.3125kHz width. (224 for downstream.) At training, it can decide to use, not use, or use at reduced bandwidth each one. It seems that AM radio stations are another interference source. -- glen
>On Fri, 18 May 2012 08:58:05 -0500, DougB wrote: > >> Hello, >> >> In every book and article aobut ADSL processing the A/D and D/A sample >> rates are given as 2.208 MHz. For a transmitter implementation at this >> sample rate you would get the conjugate symmetric image spewing out of >> your D/A since one may use up to bin 255 and there is absolutely no >> oversampling, and no analog filter in the world can cut-off that fast. >> >> Likewise for the receiver, any signal, interference, and noise above >> 1.104 MHz will fold/alias into the digitized signal because no >> anti-aliasing filter in the world will cut off fast enough to prevent >> this. >> >> So my question is: Is this really how ADSL modems are implemented?
One
>> solution would be to double the DFT size, thus oversampling the signal, >> which allows enough bandwidth for the analog filters to cut off;
however
>> I never see any reference to such techniques being used. >> >> The only other explaination is that you just take the loss and >> performance. >> >> So I was wondering if anyone can enlighten me :). > >If the transmitter and receiver ADCs are synchronized, then the signal >folding will not be nearly as much of a problem (this is obvious in the >time domain; if you grind through all the frequency-domain calculations >while dotting all the 'i's and crossing all the 't's correctly you'll >find that the signal aliases all add constructively). Interference and >noise will be an issue, but an analog filter _can_ at least cut the >amount of interference and noise, even if it can't cut all possible >interference and noise. > >It sounds like you're stuck in textbook communications engineering, which
>goes to great lengths to do optimization on signal/noise ratio, but only >mentions optimizing performance/cost in passing. I rather suspect that >the theory behind the ADSL coding is that, yes, they aren't squeezing >every last bit of capacity out of the physical channel, but that no, they
>aren't still stuck in the lab decades later, trying to come up with a way
>to implement The Best Modulation Scheme Ever for a price that consumers >are willing to pay. > >-- >My liberal friends think I'm a conservative kook. >My conservative friends think I'm a liberal kook. >Why am I not happy that they have found common ground? > >Tim Wescott, Communications, Control, Circuits & Software >http://www.wescottdesign.com >
All - Thanks for the replies - much appreciated. Tim - I think you are correct about the constructive combination of the signal and its aliased image. I thought that might be the case, so I've been re-thinking the situation and it seems to me that since you get constructive addition, then the signal power increase that you get as a consequence of the constructive addition, exactly compensates for the noise folding that will occur. So you get the same SNR as if you had wiped out the image in the first place (in theory). I'm going to work that out and simulate it just to make sure its correct. The only issues would be that any interfering signals in that alias band would be a problem, and the channel distortion between the two bins may be different, however the modems can compensate for this small distortion by reducing the bit loading in the offending bins. I have seen some modems that allow the lowpass filtered images to be transmitted and I guess that is ok as long as the spectral mask is not violated. Most modems we've seen do not transmit any image so they must oversample, which is no big deal for the transmitter. It can be a very big deal for the receiver though and at this point I'm guessing that conventional modem critically samples the signal. Thanks again for all your replies. -Doug
>On Fri, 18 May 2012 08:58:05 -0500, DougB wrote: > >> Hello, >> >> In every book and article aobut ADSL processing the A/D and D/A sample >> rates are given as 2.208 MHz. For a transmitter implementation at this >> sample rate you would get the conjugate symmetric image spewing out of >> your D/A since one may use up to bin 255 and there is absolutely no >> oversampling, and no analog filter in the world can cut-off that fast. >> >> Likewise for the receiver, any signal, interference, and noise above >> 1.104 MHz will fold/alias into the digitized signal because no >> anti-aliasing filter in the world will cut off fast enough to prevent >> this. >> >> So my question is: Is this really how ADSL modems are implemented?
One
>> solution would be to double the DFT size, thus oversampling the signal, >> which allows enough bandwidth for the analog filters to cut off;
however
>> I never see any reference to such techniques being used. >> >> The only other explaination is that you just take the loss and >> performance. >> >> So I was wondering if anyone can enlighten me :). > >If the transmitter and receiver ADCs are synchronized, then the signal >folding will not be nearly as much of a problem (this is obvious in the >time domain; if you grind through all the frequency-domain calculations >while dotting all the 'i's and crossing all the 't's correctly you'll >find that the signal aliases all add constructively). Interference and >noise will be an issue, but an analog filter _can_ at least cut the >amount of interference and noise, even if it can't cut all possible >interference and noise. > >It sounds like you're stuck in textbook communications engineering, which
>goes to great lengths to do optimization on signal/noise ratio, but only >mentions optimizing performance/cost in passing. I rather suspect that >the theory behind the ADSL coding is that, yes, they aren't squeezing >every last bit of capacity out of the physical channel, but that no, they
>aren't still stuck in the lab decades later, trying to come up with a way
>to implement The Best Modulation Scheme Ever for a price that consumers >are willing to pay. > >-- >My liberal friends think I'm a conservative kook. >My conservative friends think I'm a liberal kook. >Why am I not happy that they have found common ground? > >Tim Wescott, Communications, Control, Circuits & Software >http://www.wescottdesign.com >
All - Thanks for the replies - much appreciated. Tim - I think you are correct about the constructive combination of the signal and its aliased image. I thought that might be the case, so I've been re-thinking the situation and it seems to me that since you get constructive addition, then the signal power increase that you get as a consequence of the constructive addition, exactly compensates for the noise folding that will occur. So you get the same SNR as if you had wiped out the image in the first place (in theory). I'm going to work that out and simulate it just to make sure its correct. The only issues would be that any interfering signals in that alias band would be a problem, and the channel distortion between the two bins may be different, however the modems can compensate for this small distortion by reducing the bit loading in the offending bins. I have seen some modems that allow the lowpass filtered images to be transmitted and I guess that is ok as long as the spectral mask is not violated. Most modems we've seen do not transmit any image so they must oversample, which is no big deal for the transmitter. It can be a very big deal for the receiver though and at this point I'm guessing that conventional modem critically samples the signal. Thanks again for all your replies. -Doug
"DougB" <doug.barker@n_o_s_p_a_m.exelisinc.com> writes:

> Hello, > > In every book and article aobut ADSL processing the A/D and D/A sample > rates are given as 2.208 MHz. For a transmitter implementation at this > sample rate you would get the conjugate symmetric image spewing out of your > D/A since one may use up to bin 255 and there is absolutely no > oversampling, and no analog filter in the world can cut-off that fast. > > Likewise for the receiver, any signal, interference, and noise above 1.104 > MHz will fold/alias into the digitized signal because no anti-aliasing > filter in the world will cut off fast enough to prevent this. > > So my question is: Is this really how ADSL modems are implemented? One > solution would be to double the DFT size, thus oversampling the signal, > which allows enough bandwidth for the analog filters to cut off; however I > never see any reference to such techniques being used. > > The only other explaination is that you just take the loss and > performance. > > So I was wondering if anyone can enlighten me :).
Doug, Consider the fact that 44.1 kHz CD audio also has fairly stringent bandwidth and sample rate constraints in that a) we want to pass right up to 20 kHz and b) we want to be down 96 dB at 22.05 kHz. So you could apply the same question to CD audio. Indeed, I think you're asking a data conversion question and not a digital comms question. The truth is, most data converters these days operate via oversampling, e.g., one-bit delta-sigma (NOT sigma-delta!) ADCs and oversampling DACs, so that one really can get most or all of the digital bandwidth across without much aliasing or out-of-band components. -- Randy Yates Digital Signal Labs http://www.digitalsignallabs.com
Randy Yates  <yates@digitalsignallabs.com> wrote:

>Consider the fact that 44.1 kHz CD audio also has fairly stringent >bandwidth and sample rate constraints in that a) we want to pass right >up to 20 kHz and b) we want to be down 96 dB at 22.05 kHz. So you could >apply the same question to CD audio.
CD audio was a serious underdesign. In reality, you're better off obtaining just 18 KHz bandwidth on that channel, but specmanship prevailed, producing a lot of crummy sounding first generation CD players. (And, it has circled back around, such that current generation hardware is also pretty bad. I buy 1990-era CD players at St. Vincent de Paul because they sound better.) Steve
spope33@speedymail.org (Steve Pope) writes:

> Randy Yates <yates@digitalsignallabs.com> wrote: > >>Consider the fact that 44.1 kHz CD audio also has fairly stringent >>bandwidth and sample rate constraints in that a) we want to pass right >>up to 20 kHz and b) we want to be down 96 dB at 22.05 kHz. So you could >>apply the same question to CD audio. > > CD audio was a serious underdesign. In reality, you're better off > obtaining just 18 KHz bandwidth on that channel, but specmanship > prevailed, producing a lot of crummy sounding first generation > CD players. (And, it has circled back around, such that current > generation hardware is also pretty bad. I buy 1990-era CD players > at St. Vincent de Paul because they sound better.)
Really? Hmm, that's news to me. I've been under the impression that CD players for the last 10 years (at least) were all so good (DAC-wise) that conversion is no longer an issue. In any case, I certainly would differentiate the "media" sample rate from the "conversion" sample rate. In other words, while the media operates at 44.1 kHz, the actual conversions are done at much higher rates, digitally filtered, and continuous-time filtered with easy analog filters. This has been the case for quite some time, as far as I know. -- Randy Yates Digital Signal Labs http://www.digitalsignallabs.com