@Xenon:

Thanks for the kudos. It's actually not my first post - I used to post as "old_school", but it has been some time since I participated in the forum, and consequently, I could not remember my password...and deleting the profile would have deleted my prior posts.

Anyway...

I'm convinced it is processor related, specifically for the reason you cited about zone 2 (i.e. on the analog buss and that it has no interaction with the parallel digital processing).

Another reason that I am convinced is that I was able to synthesize three wav files, starting with white noise, and applying the delay and add approach using the time increment derived from the distance between maxima in the spectra - and what I saw makes perfect sense that this is going on and is a function of the sample rate invoked (i.e. stereo, bypass, upsample). Again, if you play the files (and I have them for anyone who might be interested in hearing them as well as the plots of their power spectra) you really cannot distinguish my re-creation (via synthesis) from the original files. Spectrally, they ARE different in the sense that while the spacing of the peaks match exactly, in the synthesized versions, the valleys are sharper; I attribute this to the fact that my synthesis was dleay and add, and therefore, does not take into account the mathematical impulse response of the 990. I could probably derive that though, and my guess is that were I to convolve the impulse response obtained from the three modes (stereo, bypass, upsample) with white noise, that their spectra would be an exact match to the data that show the issue.

I have not tried the Phono input, but I would bet the farm that in zone 2, it would not have the comb-filtering, but in the front, it would be there.

My hypothesis is this:

There is (obviously) an analog buss. Signals are taken from the analog buss and fed to an A/D, and then on to the digital engine - so far, so good. Within the digital engine the signal is manipulated however the user chooses. This signal is then fed to a D/A. Again, so far...so good.

However, my guess is that the output of the D/A feeds a summing junction on the analog buss, right before the output. This would completely explain the delay and add issue, because there is latency (processing or 'transit' time) in the digital path. Moreover, the amount of the latency is proportional to the sample rate - which explains why the timbre of the comb filtering shifts (in particular) in the upsample mode.

I suspect that there is a switch at the summing junction (whether an electronic switch or a relay) that is supposed to disconnect the output of the D/A before the summing junction, thereby ensuring that there really is a true bypass. After all, bypass is supposed to bypass all processing circuitry, but the fact that comb filtering happens even in the bypass mode tells me that analog and digital signals are being summed. What puzzles me is why the dolby mode(s) don't exhibit the comb-filtering, but frankly, I don't really know a lot about those codecs.

Anyway, it seems to me that the way the analog and digital traffic should be managed is thusly:

*Assuming* the analog signal is always fed to the digital section, there should be a two-position 'switch' after the D/A; in one position, the output of the D/A is fed to the summing junction (digital modes). In this switch position, a second switch is in a complementary state. That is, when any digital mode is selected (i.e all but BYPASS) the analog signal is disconnected from the summing junction.

Using this architecture, it would be impossible for 'delay and add' to occur, because the analog signal (from which the digital signal is derived...and with latency) could not be summed with the delayed digital signal.

Conversely, in bypass mode, the two switches each have to change state to their respective complementary states (complementary to the states used in digital modes). That is, in bypass, the signal that is rattling around in the digital section is effectively disconnected from the summing junction, and the only thing that the junction can add is the analog signal on the analog buss to "zero" coming from the digital section.

As an electrical engineer, I know that when digital systems are created, truth tables are employed to determine the combinational or sequential logic. However, sometimes, there are conditions called 'don't care' that stem from the fact that a particular combination of inputs will never arise, or if they do, the designer 'doesn't care' what stse the device is in - this is done to keep costs low, but in 'critical' systems (i.e. those in which human life hangs in the balance) don't care's are very carefully managed - don't care's become 'you better know what this will do when these particular inputs show up and you had better act accordingly'.

So, if there are in fact these switches present (i.e. my guess at the architecture is correct) and if they are addressable electronically, then my guess is that someone missed how the don't care's were assigned. That's actually my hope, because this would seem to be something easily remedied in a firmware update (i.e. proper management of the switches, if indeed this is how the architecture works...which again...I don't know...I'm just looking at what the data are telling me, as well as the results of my own experiments to get a handle on this).

However, if the analog and digital (post D/A) streams are always summed (i.e. none of these 'switches' or routing provisions exist), then it would be impossible - mathematically and physically - to prevent a delay and add issue from occurring; this would be a much bigger problem for those who wish to listen to analog inputs in the main zone. Again, if all you care about is zone 2, then this is not an issue...

...but...there are analog inputs...and...

...there is a 'bypass' mode which is supposed to be purely analog...but how can comb filtering happen if the bypass is really working?

All of the math point to a delay and add scenario. Maybe later I can post the delay times that I calculated.

One other thing...I am not grousing about the delay - I could care less about that, because merely delaying a signal will not produce comb-filtering. However, delaying it and adding it will do this. In fact, this is the heart of flanging (Think of the bridge in the Doobie Brothers' "Listen to the Music" or the Small Faces' "Itchykoo Park" and you'll know what I am talking about). The only difference between comb filtering and flanging is that the amount of delay time is variable, and thus, as one sweeps the delay time, you get that inside-out sound, but it changes in timbre as the flanging goes on; stop varying the delay and you get a comb filter once more.

I do use it for HT; I take the optical from the DVR and to be honest, I have never had to experiemnt with the lip-sync delay. However, I don't see this as associated to the problem...but I could be wrong. Still, it seems unlikely, because there is no delay on the analog buss - all the delay that takes place does so in the A/D --> PROCESSING --> D/A path...because discrete sampling is going on.

Phew...time for a drink. Again, if anyone out there wants to hear / see the files in question, let me know...but a simple experiment would be to set the tuner to a non-station (to generate something close to white noise) and then cycle the modes (stereo, bypass, upsample, dolby, dolby vs, dts etc) and listen carefully to see if your 990 exhibits the same issue.

This could simply be a defect in my unit (bummer for me, but at least otehrs don't have to deal with it)...or it could be an oversight (back to the overlooked don't care's thing...) ...or it could be a design ovesight for which there is no fix. Again, I have no '2nd 990' lying around against which I could compare this particular 990, so I don't know what the answer is (obviously...or I would not have posted this).

Thanks...

Mark