Keef wrote on Wed, 22 December 2004 18:00 |
Here's a ex BBC engineers opinion
The simple fact is that the vast majority of converters don't do what the theory calls on them to do when operating at 44.1 or 48kHz. The anti-alias/reconstrution filters have an audible impact onthe pass band when they shouldn't. They introduce amplitude ripples and horrendous phase distortions....
Hugh
-------------------- Technical Editor, Sound On Sound
|
"The simple fact is that the vast majority of converters don't do what the theory calls on them to do when operating at 44.1 or 48kHz. The anti-alias/Reconstruction filters have an audible impact onthe pass band when they shouldn't. They introduce amplitude ripples and horrendous phase distortions. And if they are designed to minimise these aspects, the transition band is insufficiently steep and the stop band insufficiently attenuated, resulting in aliasing distortions."That was true 15 years ago. The AD converters today are operating at over sampled rate and the anti aliasing analog filters phase problem is NOT an issue any longer.
DA’s also operate at up sampled rates and the analog reconstruction filter is no longer a problem. Up sampling is mainly done FOR THAT REASON, to overcome phase problems. When is the last time you saw a DA with no up sampling?
When it comes to the digital up sampling filters (for DA), if they are well designed, there is minimal ripple. The FIR’s interpolating yield linear phase, thus NO phase problems.
When it comes to the digital decimation filters (for AD), if they are well designed, there is minimal ripple. The FIR’s decimators yield linear phase, thus NO phase problems.
"They don't work properly at 88.2 or 96kHz either, but ther is so little audio signal energy close to the turnover points that the problems don't manifest."So that is also incorrect. If you want to find phase problems near 20KHz, look at the mics and the speakers…
As always, inthe affordable end of the market, quality is limited by cheap parts, cut corners, and poor design implementations. Nothing new in that -- the same problems affect analogue products too.
hugh
Technical Editor, Sound On Sound That part is correct.
"It is certainly true that some instruments generate ultrasonic energy -- trumpets, some percussion, string sections etc. It is also true that some of the these ultrasonic components can interact in the air to produce audible intermodulation products. I think this is one reason why miking a string section from a reasonable distance always produces a richer, more pleasing sound that close miking each instrument and mixing in a desk!
Also, very few microphones have a response that extends significantly above 25kHz or so, and the same for loudspeakers. So in most cases, the mic is not capturing ultrasonic energy even if it is there, and neither can the speaker reproduce it."I have been saying that for a long time. I am glad it is being heard. First, lets be very clear. It takes mics AND speakers AND ears to respond to higher frequencies. Second, With say 88.2KHz you have OVER 40KHZ AUDIO! That is more than is needed.
But, the high end roll off in both cases is gentle -- 6dB octave, typically -- which means there is little phase distortion.A musical instrument is not a filter. If it makes say 25KHz harmonic, it is by definition zero phase! Do we need to capture it? Not if we do not hear it! Having it removed does not impact the sound. So 50KHz trumpet energy is of zero value (for humans and even for dogs). Again, your comment
the anti-alias/Reconstruction filters have an audible impact on the pass band when they shouldn't. is wrong in the environment of the last 10-15 years of audio. You obviously did not read my papers. Look at “Sampling Theory” and also at “Sampling, Oversampling, Imaging and Aliasing” on my web under support.
"Contrast that with digital converters with brickwall rol-offs operating at 44.1 or 48kHz sample rates. These inherently cause horrendous amounts of phase distortion around the turnover freuqnecy, and that, I think, it what our ears pick up on as 'the digital sound' that many don't like. Move the sampling rate up to 96kHz, and while the phase distortion still happens, it is now way outside the hearing range and so the sound appears to have improved."Once again, a same wrong story repeated here. A well done, or even a poorly done digital decimation, and also digital up sampler, when based on FIR (not IIR) yields ZERO PHASE SHIFT!
A poorly done filter will have ripple. But instead of saying horrendous amounts of phase distortion around the turnover frequency, you should say understand it has ZERO PHASE DISTORTIONS.
Again, the analog circuits are operating way up there (frequency wise). Modern AD’s front end operates at 64-512fs (2.8Mh to way over 22MHz)! Nyquist is so high that the analog filters today are rarely above 3 poles – no horrendous phase shifts anywhere, not even in the MHz. A 3 pole filter yield 135 degrees at cutoff, and only a few degrees at an octave bellow cutoff. So if your cutoff is at say 50KHz, your deviation from linear phase is under already control. Disagree? So set your filter at 100KHz or 500KHz!
Most DA’s operate at significant up sampling rate and the filter corner frequencies is also way above 20KHz. Similar story to the AD – NO horrendous phase shifts.
Why are you talking about phase shifts resulting from filter corners at such high frequencies when your microphone and speaker have a corner at 20KHz or so?
I find the spread of this sort of information to be very disturbing. Plainly wrong facts, mixed with a couple of correct statments, when stated with such authority, are the reason even large companies chose to ignore engineering and scientific fundamentals and go for the outrageous 192KHz hype.
Also, while promoting everything British as good (Prism, DCs, Mr. Craven, Neve Counsel) and even putting down some non British makers, lets not forget that it was a DCS paper that is responsible to propagating much of that 192KHz BS.
Regards
Dan Lavry
http://lavryengineering.com“In a time of deceit, telling the truth is a revolutionary act.”