Jittered was the trouble often involved with the SAA7220, despite PCM chips are said not too much affected by jitter ! Or are they with oversampled material ?
Anyway, my NOS TDA1541 based dac is glueing to the definition to the Lampie's one and it has in subjective listening tests grounded many branded DAC SG based costing few grands ! IT is DC coupled.
Anyway, my NOS TDA1541 based dac is glueing to the definition to the Lampie's one and it has in subjective listening tests grounded many branded DAC SG based costing few grands ! IT is DC coupled.
Last edited:
So you don't like ABX because of the results aren't what you preferred but you do subjective listening comparisons because results are what you preferred. Got it.
It can be done, though, even with some degree of group delay equalization if you really want to.
@Lampie,
You strike me as a pretty smart fellow, I'll bet you did just fine in school.
OS, of course, involves low-pass filtering of the image-bands, leaving only the baseband, plus some residual image-bands. Removing the image-bands is a requirement for fully reconstructing the signal in accordance with the sampling theorem. NOS does not filter the image bands, well, except for the SINC envelope produced by a converter's zero-order-hold operation. Therefore, NOS does not properly reconstruct the signal according to the sampling theorem.
This touches on the original question posed by this thread, and it also reflects your point, I believe. Which is that since the image-bands are ALL ultrasonic, the signal from NOS DACs is being band-limited (although not completely) by most loudspeaker's tweeters, and then by the ears of listeners.
So, then, why do the two sound different? There must be an technical explanation which fully makes sense.
Last edited:
This seems interesting, Marcel. Is the resulting filter very complex? Can you provide a link or two for reading?
Like Abraxalito (Richard), I have obtained some interesting levels of image suppression via 7th order Cauer filters, but obtained no more than about -20dB worth at 25KHz while also keeping 20KHz EQ'd flat - in simulations.
Last edited:
I remember seeing a smart solution on TNT audio long time ago (Decimation). Simply explained it is just a S/H on the output of the DAC chip.
Naturally the NOS version has a roll off at higher frequencies (i did not see this as technical inferior btw).
This can be one reason why we haer a difference but that is not really noticable in my opinion.
The main problem (according to Lampie519) is the bad FIR's used in general because of lack of recources at the time (chip space, costs etc.)
I remember the first CD player coming onto the market (Sony. Philips and all others beeing a copy of those or licensed).
Now these were really expensive and already "High End" at the time (marketing wise). Now only Philips was making a FIR at the time as Sony claimed 16bit already. So there was no alternative other then purchasing chips from them and later BB and NPC, what have you.
These companies were not manufacturers of High End audio but want to sell "sand", that's all. If a company wanted to have something better they had (and still have) to make their own chips. Now that is something you just can not startup and hope to make some kind of profit and programmable chips where not that powerfull yet.
Only now we have these available and you see that finally there are FIR's available that can do the job !
Naturally the NOS version has a roll off at higher frequencies (i did not see this as technical inferior btw).
This can be one reason why we haer a difference but that is not really noticable in my opinion.
The main problem (according to Lampie519) is the bad FIR's used in general because of lack of recources at the time (chip space, costs etc.)
I remember the first CD player coming onto the market (Sony. Philips and all others beeing a copy of those or licensed).
Now these were really expensive and already "High End" at the time (marketing wise). Now only Philips was making a FIR at the time as Sony claimed 16bit already. So there was no alternative other then purchasing chips from them and later BB and NPC, what have you.
These companies were not manufacturers of High End audio but want to sell "sand", that's all. If a company wanted to have something better they had (and still have) to make their own chips. Now that is something you just can not startup and hope to make some kind of profit and programmable chips where not that powerfull yet.
Only now we have these available and you see that finally there are FIR's available that can do the job !
Last edited:
Depends on the type of non-oversampling DAC. The non-oversampling DACs of the 1980's and before all had analogue reconstruction filters. Non-oversampling DACs with no filter at all except a zeroth order hold came into fashion much later. Of course any practical reconstruction filter is imperfect and doesn't fully do what it is supposed to do according to the sampling theorem, which is fortunate, as any causal ideal low-pass filter has an infinite delay.
No, except for general references about filter design (Anatol I. Zverev's Handbook of filter synthesis) and group delay equalization (DeVerl S. Humpherys, "Equiripple network approximations using iteration techniques", Proceedings of the National Electronics Conference, vol. 20, 1964). Still, I remember that all CD players from the early 1980's except those from Philips had steep analogue reconstruction filters. Most of them had minimum-phase filters, but some of them featured some degree of group delay equalization. No idea how well they suppressed images.
He's right, as usual. In fact any sampling system with imperfect suppression of aliases and/or images must have a time-variant impulse response.
Linear time-invariant systems are incapable of producing frequencies that aren't there in their input signal or input signals, so the fact that there are images and aliases shows that the system is either non-linear or time-variant or both. As long as nothing clips, folds, quantizes or adds noise, superposition still holds, so the sampling system is in fact linear. It must therefore be time-variant if it produces aliases and/or images.
You can see that quite clearly when you look at just a bare sampler with no filter at all. As an example, suppose it samples every second, at t = ..., -3 s, -2 s, -1 s, 0, 1 s, 2 s, 3 s,...
When the input signal is 1 between t = -0.2 s and t = +0.2 s and 0 everywhere else, the sample for t = 0 will be 1 and all other samples will be 0.
Now shift the input signal by half a second, so it is 1 between t = +0.3 s and t = +0.7 s. The sampler will then only produce 0 samples, so the 1 sample has disappeared. If the sampler were time-invariant, a time shift of the input signal should result in an output signal that's equally shifted in time, but otherwise looks the same as before the shift - so the 1 sample should shift by half a second instead of disappearing.
With a perfect anti-aliasing filter and a perfect reconstruction filter, the signal coming out of the reconstruction filter is a perfect reconstruction of the output signal of the anti-aliasing filter, and it remains so when the input signal to the anti-aliasing filter gets shifted.
Exactly.
So all in all, assuming proper anti-alias filtering at the recording side, the ultrasonic images that you get without good anti-imaging filtering make the signal chain time-variant, but the part below the Nyquist frequency is reproduced properly anyway.
Last edited:
That would be all known (to me) NOS DACs. A reconstruction filter which fits within a 16bit error budget would be a more than 20th order Chebyshev(*), assuming a 0.1dB flat passband to 20kHz and stopband beginning at 24.1kHz.
Zanden's been mentioned but as far as I'm aware their very complex reconstruction filter isn't intended to achieve 16bit stopband performance rather improved phase response.
(*) An elliptic or Cauer filter would be able to achieve the target response with fewer stages than Chebyshev.
Last edited:
Oversampling DACs also don't necessarily meet your criterion. I once ran into some strange distortion measurement results using a CD player as a signal source and ended up measuring the imaging products at 44.1 kHz +/- f of an Aristona CD1380 CD player, which is a standard oversampling TDA1541 design. The test signals were 10 dB below full scale, unless otherwise noted. The results were, in dB with respect to the desired signal:
f = 100 Hz: below my measuring threshold
f = 1 kHz: -62.47 dBc at 43.1 kHz, -63.95 dBc at 45.1 kHz, -77.9 dBc at 87.2 kHz, -78.27 dBc at 89.2 kHz, below the -83.12 dBc measuring threshold at 131.3 kHz, 133.3 kHz, 175.4 kHz and 177.4 kHz (it surprised me that the levels at 175.4 kHz and 177.4 kHz were so low, but they were)
f = 1 kHz, but 0 dBFS: -62.37 dBc at 43.1 kHz, -63 dBc at 45.1 kHz
f = 10 kHz: -71.63 dBc at 34.1 kHz, -68.16 dBc at 54.1 kHz
f = 100 Hz: below my measuring threshold
f = 1 kHz: -62.47 dBc at 43.1 kHz, -63.95 dBc at 45.1 kHz, -77.9 dBc at 87.2 kHz, -78.27 dBc at 89.2 kHz, below the -83.12 dBc measuring threshold at 131.3 kHz, 133.3 kHz, 175.4 kHz and 177.4 kHz (it surprised me that the levels at 175.4 kHz and 177.4 kHz were so low, but they were)
f = 1 kHz, but 0 dBFS: -62.37 dBc at 43.1 kHz, -63 dBc at 45.1 kHz
f = 10 kHz: -71.63 dBc at 34.1 kHz, -68.16 dBc at 54.1 kHz
Thanks, for clarifying what Bruno was saying in the presentation which I had seen. So, if I now correctly understand, with NOS, the relative phase/timing between the sample moment and the input signal causes the DAC impulse-response to be time-variant. However, shouldn't this only affect the ADC, but not the DAC, which can only accept whatever samples are given it, and can not do anything about samples which may be incorrect, or missing entirely?
Looking beyond my remaining confusion on the above question, what impact, then, does a time-variant DAC impulse response have on the baseband signal? Shouldn't there be some sort of phase-modulation of the baseband after all?
Last edited:
Good question. My comments apply to the entire chain, anti-aliasing filter (if any), ADC (effects of quantization ignored for simplicity), DAC, reconstruction filter (if any). With no or inadequate filters, this whole chain is time variant. That is due to the sampling, which inevitably takes place in the ADC, so you are right that it is the ADC's rather than the DAC's fault. Solving it requires either getting rid of the sampling, which would mean switching to analogue, or using good filters on both the ADC and the DAC side.
So all in all, you need a reconstruction filter to get rid of the time variance of the signal chain that's caused by the ADC, and it only eliminates the time variance when the recording was made with a decent anti-aliasing filter.
Spectrally, the effect of the time variance of the signal chain that you have with an anti-aliasing filter but no reconstruction filter, is the whole bunch of spectral copies of the signal around multiples of the sample rate. The part up to the Nyquist frequency is not affected.
Conversely, with a reconstruction filter but no anti-aliasing filter, the effect of the time variance is aliasing and the part of the spectrum up to the Nyquist frequency can be affected.
I think I'm explaining things that everyone has known for decades in terminology that isn't often used
Last edited:
I imagine that the various effects that you can describe - and Ken's list - can be modelled and expressed as time domain effects within the audio band? At least in terms of approximate scale if not in detail. Then it could be compared in terms of scale with e.g. the effects that Prof Hawksford described in I/V circuits where slew rate was inadequate for high-frequency pulses and caused dynamic variations in loop gain leading to time domain jitter.
Hans,
We are almost at that step. Assuming that our list is essentially complete, we should now begin the process of, hopefully, being able to eliminate some of the suspected reasons by process of deductive reasoning. I doubt that we be able to coordinate and conduct the sort of scientific investigation we would like to, given the practical difficulties presented by most of us being hobbyists working in our basements or garages, with minimal resources. However, we will attempt to press forward, just the same.
I think we can begin submitting logical arguments to the thread, for review by everyone, on why certain of the suspected reasons may be logically eliminated, without spending the time and energy to obtain proof by conducting an actual experiment on every one of the candidate reasons. That's just my present thinking, which may not be sufficiently imaginative, or may be just plain faulty. I'm certainly open to hearing a better, or a more efficient approach to the investigation.