The Great DAC Mystery
This plethora of DAC’s phenomenon was such a mystery to me for 20 years. How can measurements be so incredible, yet many continue to prefer DACs that don’t measure so well. And almost everyone agrees they sound different (significantly in many cases). Why don’t the good ones sound the same. ASR are right in many ways - measured performance is important - but a pure focus on measured performance is completely wrong in my experience (using my ears). And here is my explanation of why!
Finally I believe I have stumbled upon a huge part of the problem with DAC technology. Of course it all stems from the inadequacy of measurements and even the technical instruments (audio precision) used to conduct those measurements - this is all at the root of why measurements are failing to be a reliable tool to select a DAC. There’s more though - if you read on please consider my reasoning and give my solutions a try - you may be surprised at the audible improvements that can be easily obtained.
There are a few things that hint at the problem of playing Redbook 44.1 source music:
1) R-2R DACs - why the resurgence?
2) Vinyl resurgence
3) The brick wall vs smooth, linear vs minimum phase debate: M-scaler, HQ player, FPGA XIlNIX proprietary programming, a plethora of filters.
4) HQplayer, PGGB and precursors like SACD - why is DSD still around and why do some people prefer it to PCM?
First let’s recognize that: All of these things can’t possibly be just coincidence!
So what is the underlying ROOT CAUSE:
Passband Ripple (‘equiripple’ to be precise)
1) All DAC’s are basically Sigma Delta DACs (which make up 99.99% apart from the recent handful but growing number of audiophiles with R-2R DAC’s). These Sigma Delta DACs ALL rely on upsampling to work - the final conversion is 1 bit or parallel 1 bit converters.
2) All upsampling DAC’s will take Redbook 44.1 (the vast amount of available music is in this format) and upsample (usually 8x initially but often higher) using short tap filters with low latency that have excellent specs but universally create a tiny but non-negligible passband sinusoidal ripple (it isn’t supposed to be audible).
MATH FACT: A sinusoidal ripple in the passband (what range of audio frequencies are presented to the listener) is equivalent to a pre and post-echo in the time domain (the signal you hear coming out the speakers)
The MANIFESTATION: Digital glare, harshness and a poor soundstage (the harshness is sometimes confused with accuracy - it is actually distortion - but not distortion that you can measure with an analyzer, as it is just like a reflection - it contains a reflection of the entire audio signal displaced in time at low amplitude ). Types of filters will have different forms of passband ripple - these lead to slight differences in the distortion (pre and post-echoes can occur at different times before and after the true audio signal - some time differences being more audible than others).
The SOLUTION:
There are three options
1)NOS with an R-2R DAC (can still suffer from aliasing which can create IMD in passband and the final filter can also create passband ripple)
2) upsample using a PC at such very high precision as to reduce passband ripple to inaudible levels (upsample can be to PCM or DSD but it might as a well be DSD as most DAC’s convert PCM to DSD anyway, only an R-2R DAC would be best fed upsampled PCM)
3) Vinyl - for the most part vinyl does not suffer from these issues at all but of course you get pops, cracks, surface noise, less channel separation, variability of pressing quality, and, if competing with digital; the need for very high end TT, phono-pre, cartridge, careful setup etc.
Anyway, please read carefully and think about the above with an open mind. Passband ripple is the elephant in the room that nobody talks about. Remember that very little if any testing has been done on our ability to hear pre-echoes however, anecdotally, all speaker builders recognize that a sharp baffle edge causes edge diffraction which is recognized as being audibly detrimental to the sound (and affects stereo imaging) Hence all the narrow speakers and exotic attempts to keep midrange and tweeter baffle width very small (think of all those countless big highly regarded audiophile three ways that are big on the bottom but narrow at the top)
It’s been a while, I thought I’d share this. No need to argue about this. I will offer clarifications but those who don’t get it or buy any of this will just miss an opportunity for better sound - I’d rather not argue with you. And, for those who will conflate pre-echo or post-echo with pre-ringing or post-ringing - I am NOT talking about ringing at all - the echoes I refer to are complete true echoes of the entire audio signal - equivalent to and analogous to a reflection off a wall.
I've always felt that software upsampling yields better results (to my ears at least) than hardware upsampling. I've had a number of DSD DACs (PS Audio--I was a beta tester for the original PerfectWave, EMM Labs, Playback Designs) in my own system and moved on to R2R designs with software upsampling (HQP). Just more musical IMHO.
|
An interesting article about DAC filters and ripple: https://addictedtoaudio.com.au/blogs/how-to/how-to-pick-the-best-filter-setting-for-your-dac I remember reading a Stereophile article way back in the old days. One of the reviewers had got into his possession a DAC of some sort that would let him write and apply his own filters. He was pretty sure that ringing filters were the culprit for what he was hearing that he didn’t like about digital. His conclusion after writing and listening to some horribly ripply filters was... filters weren’t the problem. He couldn’t make the DACs sound any better or worse to his ears by eliminating ripple. I bought a NOS, filterless dac with an old chip because I wanted to hear it for myself. I couldn’t hear any improvement like I was expecting from that either. I think most of what people are hearing and referring to as things like microdynamics and timbre are happening on much longer time scales than any problems with timing that digital is introducing, or pretty much any electronic gear such as amps or pre-amps. However, anything that alters frequency response even just a little bit can result in all sorts of unintuitive perceptions about timing, I’ve tested my hearing on timing issues, creating signals that I thought would audibly reveal timing issues with slower bitrates compared to higher. Measurements showed the higher bitrate effect was actually coming through the speaker, but I definitely couldn’t hear the difference. That’s when I realized I was making timing effects that occur faster than 1/20,000 of a second. I can’t hear that high, so I can’t hear the timing that high either. |
@asctim What you refer to seems to be the conventional wisdom and accounts for 99% of the literature. However this Gibbs ringing is not audible to the majority of people as the frequency is at the transition frequency of the filter (the point where you have the filter acting) - and this frequency is almost always above 20KHz. This is a fact and therefore makes this pre-ringing or post-ringing irrelevant as far as our hearing is concerned. I don’t believe we hear it at all. We do hear of course the affect of minimum phase filters because they alter phase relationships and therefore the timbre. (Some of us might notice a very smooth filter that rolls off at 15KHz - those younger listeners with full frequency range hearing) I propose that what actually matters is changes in phase (timbre) and the passband ripple. The passband ripple creates two echos ( a complete reflection of the entire audible signal ) - a pre-echo and a post echo. By changing the filter type and it’s steepness: the most audible impact will be a combination of two things: 1) the way minimum phase will change relative phase and therefore the timbre compared to the theoretical optimal of linear phase (which preserves phase/timbre) 2) the change in the ripple within the passband - which affects both the echo timing and amplitude. High quality (heavy processor overhead) HQP filters allow for a passband ripple to be incredibly small (many orders of magnitude smaller than a DAC FPGA) so upsampling to DSD with HQP will eliminate the pre and post-echoes found in all regular Sigma Delta DAC’s (due to their limited processing power) An alternative is an R2R NOS DAC used in NOS mode (as this will not suffer from sinusoidal passband ripple with practically any correctly designed analog output filter)
|
I did some reading on the subject. I read the echo is under 1ms offset and -60 dB or more below the main signal. That’s really quiet, and not a lot of timing offset. The thing to show is if under any conditions anyone can pick up on that, maybe not even hearing it directly, but sensing it somehow changing their perception of the sound. Perhaps some people can. I suspect I can’t. I’ve tried mixing echoes of various types in to recordings to see what effects it might create. I quit hearing any notable change by -35 dB. Maybe even -30 or -25, depending on the type of echo and the timing. If you want to hear something that really stands out, use a pitch shifter to slightly de-tune the echo. That’s a really bad sounding effect. Even that one disappears for me long before -60 dB. |
Yes I agree. Typical DAC 2x upsampling filter will have 0.7 msec pre and post echoes of amplitude -67db of main signal. It’s incredibly small but certainly in theory, this is in the actual humanly audible range, and considering the pre-echo will not be masked. I will explain below why I believe we can hear subjectively this tiny echo as harshness and as telling our brain the sound came from the speaker (will reduce quality of stereo image). We are very good at locating sounds above 6kHz using the interaural difference in sound levels at each ear (our head being responsible for heavily attenuating higher frequency sounds coming from our right from reaching the left ear and vice-versa). The echo, although very small, therefore causes confusion for proper location of the stereo image. It gets worse; this echo at 0.7 msec is very close to the actual interaural time difference of hearing a sound at each ear from extreme right or left - so even sounds between 300 and around 2KHz may not image as well (this frequency range is where human hearing tends to rely on time arrivals at each ear to sense sound location). Now worst of all, this echo is a true echo - it is an entire reflection of the audio signal - it perfectly correlates to the music or vocals! So although very small, our hearing is well developed to detect it - especially as that is precisely what our ears/brain are listening for in order to detect sound location from left to right. The above is also why speakers with baffles less than 9 inches will image very well and those speakers with baffles of 12 or more inches no longer “disappear” and no longer image so effectively (due to our ears detecting the echo from the baffle edge diffraction, being only perceived once it exceeds about .6 msec in delay). This phenomenon is well understood by speaker designers - the manifestation being the large number of speakers that have a narrow baffle just for the mid range or tweeter.
|
