Between 30 kHz and 40 kHz is enough IMO. Amplifiers that go higher up in frequency would theoretically help in reducing the group delay distortion at the upper hearing frequencies. Be aware that it can't improve the situation it can only help so much that it doesn't add further group-delay distortion to what is already there and that is coming form different sources in the whole production/reproduction chain.
In an active system for instance it would be possible to correct for the part that is under one selfs control (i.e. amp and tweeter) by the use of phase-equalising techniques. That way it would be possible to make a system that has flat group-delay even with an amp that has an upper cutoff frequency of 20 kHz. No need for amps that could be used as broadcast transmitters.
Regards
Charles
In an active system for instance it would be possible to correct for the part that is under one selfs control (i.e. amp and tweeter) by the use of phase-equalising techniques. That way it would be possible to make a system that has flat group-delay even with an amp that has an upper cutoff frequency of 20 kHz. No need for amps that could be used as broadcast transmitters.
Regards
Charles
Last edited:
Well, I have mixed feelings about this...
Granted, I can't hear a 20kHz sine wave.
OTOH our ears don't work in the frequency domain, but in the time domain. Remember that the frequency domain is only a theoretical, abstract construction to analyse real, time-based phenomena.
And in the time domain, a bandwidth limitation does definitely change (ever so slightly) the wave form...
Now can we really hear that ?
Granted, I can't hear a 20kHz sine wave.
OTOH our ears don't work in the frequency domain, but in the time domain. Remember that the frequency domain is only a theoretical, abstract construction to analyse real, time-based phenomena.
And in the time domain, a bandwidth limitation does definitely change (ever so slightly) the wave form...
Now can we really hear that ?
In my youth (late teens, up to 22 or so) I could hear up to a 21k sinusoid, and I could readily hear the 19kHz pilot tone on my FM receiver as I switched between it's deep and shallow filter (Technics ST3500) and I'm sure we could all readily hear the audible sound from a CRT tube (16.5k or so)
Now my hearing response is comparatively seriously reduced and I have tinitus.
But I still seldom (if ever) hear (CD) reproduced audio that sounds like real life for things like: sibilance, cymbals, gongs, soprano/tenor head voices, the residual 'breathing' of a violin (rosin across the string) when high frequency notes are played quietly), trumpet rasp.
I'm of the belief that these are primarily down to what is happening in the 16-25k range of the frequency spectrum and have a similar feeling to David dP that it's actually because our ears do processing in the time domain and these high-frequency components are important even to my fading hearing system
...Apologies for trying to provide a technical justification without documentary evidence, but there's too much of a general mis-match between what I hear in real life and what I hear form the reproduction chain, and it's trying to work out why
Now my hearing response is comparatively seriously reduced and I have tinitus.
But I still seldom (if ever) hear (CD) reproduced audio that sounds like real life for things like: sibilance, cymbals, gongs, soprano/tenor head voices, the residual 'breathing' of a violin (rosin across the string) when high frequency notes are played quietly), trumpet rasp.
I'm of the belief that these are primarily down to what is happening in the 16-25k range of the frequency spectrum and have a similar feeling to David dP that it's actually because our ears do processing in the time domain and these high-frequency components are important even to my fading hearing system
...Apologies for trying to provide a technical justification without documentary evidence, but there's too much of a general mis-match between what I hear in real life and what I hear form the reproduction chain, and it's trying to work out why
Hello ChrisPa
Could you describe your hifi system, from front end to amps to speakers.
Regards
Arthur
I could, but why?
I'm not talking about my hifi system. I'm talking about all the various hifi I've heard, and (I believe) I know what I aspire to - what I've heard that comes closest to that illusion of reality and what the shortcomings of my hfi are and what needs replacing. And this in comparison to real life (amongst other music, I sing tenor in chamber choirs, frequently standing adjacent to orchestras)
Why am I on this thread? Because I have an inherent belief that all the design, measured and observed criteria that Bruno has included/considered within the (UcD and) ncore are an inherent part of a transparent reproduction system, helping me get closer to that illusion - look at the distortion vs freq, distortion vs amplitude, output impedence vs freq, and the lack of variation in frequency response with load. And because this summer (very wet in the UK, so an approximate description) was spent landscaping the garden so I never made time to build some UcD 400 amps
So I'm not sure what the benefit of me describing my existing hifi would be.
"I haven't yet found a car that accellerates fast enough" and "I drive a Chevrolet Matiz" may both be true, but the latter statement and me describing my Matiz in more detail doesn't negate or change the first statement and doesn't help me get to where I want to be.
(BTW, I don't drive a Matiz, and my hifi isn't a Matiz)
Ask instead what I have heard that gets closer to that illusion and/or how I think I should get there. My power amps will be nc400s
... but I think we might be taking this thread off-topic
Actually, not quite. Nerves work on pulse density modulation, with pulse repetition rates up to one or two kilohertz, so from that one can infer a few things about phase audibility:
[1] Phase response is most audible at lower frequencies, and indeed that is the case. Flat phase response correlates well with more realistic presentations of percussion instruments, no surprise since snare drums have an especially fast leading edge on the waveform.
[2] Phase response is not audible on frequencies above a few hundred hertz, shown in a listening test at BAF 2007 or 2008.
The cochlea is a series of tuned filters, with cancellation between adjacent filters to provide sharper frequency discrimination, so the frequency domain is very real to human perception. Consider how we define notes: middle C, high E, etc. They are not called by the wave shape, which was mostly unknown until high speed visualisation showed it in the 20th Century.
Moreover, one cannot speak of the frequency domain as an abstract construction over a real phenomenon. It's every bit as real as the time domain, merely less obvious to the eye because it doesn't emerge as a series of points on an oscilloscope. One could say waveforms are an abstract construction of the real time frequency domain since the former were quite unknown before people could see audio.
Perhaps not. I'm willing to take bets on tweeter nonlinearities interacting with Gibbs ringing, however.
I think I should deconstruct the argument concerning audibility of high frequencies a bit. If a listener can't hear a pure tone of 20kHz, does that logically imply that spectral content above 20kHz is inaudible? Not by itself. There is the hidden assumption that the ear is a linear system. If it were, the assertion would be incontrovertibly true. However, this assumption is clearly false. Ears aren't spectrum analysers. The hair cells are zero-crossing detectors. The cochlea is an adaptive filter. It is not a linear system. The inaudibility of pure HF tones does not imply that >20kHz spectral content is necessarily inaudible.
The question therefore becomes: is there a frequency where we can safely say that the ear has thrown in the towel? Only listening tests (I mean good ones, not acoustic wine tasting sessions) can answer this. Since it remains a hot topic in some circles, the tests keep coming and the results are consistent: you can insert a 20kHz low-pass filter undetectably provided it is not too steep. Amazingly the sort of filter we use in AD/DA is still gentle enough. What matters is that the filter is free from pre-echos and other nonidealities (which most AD/DA filters aren't ) whilst remaining linear phase. On the other hand, a filter with a 100Hz transition band is very, very audible.
So the discussion whether >20kHz content is audible has been seriously fogged by the question whether certain avoidable shortcomings of digital brick-wall filters are audible. Because some commonly used filters are all too audible people have gone on a sampling rat race that still hasn't abated. DXD anyone? As a sanity check I would suggest that people try a seriously good sample rate conversion software like SaRaCon. Take a 96kHz file, take it down to 44.1kHz and then back to 96kHz to eliminate systematic errors. Now play both back and be ready to be amazed how transparent the process can be.
The upshot of all of this is: do not draw conclusions from pure tones, but on the other hand the 20kHz limit is pretty safe unless you deliberately push your luck.
Now, what does that mean for amplifiers?
1) Even if the audibility of >20kHz content may be subject to discussion, the fact that <20kHz content is much more audible than >20kHz content is not controversial anywhere. I've never heard it claimed that above 20kHz is more important than below, even by the most seriously deluded woo-woos.
That means that trying to optimize for supersonic performance is always a losing battle because even if 1dB of performance gained outside the audio band only incurs a 1dB performance loss inside the audio band, the performance loss is always much, much more readily audible than the supposed improvement. In fact, the noise shaping theorem says that the trade-off is much, much worse than a dB for a dB.
2) Limited amplifier bandwidth can only be a liability if the roll-off is very steep or the phase response is not linear. A Bessel (Thomson) filter has linear phase inside the bandwidth which equates only to a delay. Ncore has a mainly 1st order roll-off, which is Bessel by definition.
3) If you want to know something sonically meaningful about the high-frequency distortion of an amp, do a high-frequency IMD test. If the amplifier survives that, you can stop worrying about distortion above 20kHz.
The question therefore becomes: is there a frequency where we can safely say that the ear has thrown in the towel? Only listening tests (I mean good ones, not acoustic wine tasting sessions) can answer this. Since it remains a hot topic in some circles, the tests keep coming and the results are consistent: you can insert a 20kHz low-pass filter undetectably provided it is not too steep. Amazingly the sort of filter we use in AD/DA is still gentle enough. What matters is that the filter is free from pre-echos and other nonidealities (which most AD/DA filters aren't
) whilst remaining linear phase. On the other hand, a filter with a 100Hz transition band is very, very audible.So the discussion whether >20kHz content is audible has been seriously fogged by the question whether certain avoidable shortcomings of digital brick-wall filters are audible. Because some commonly used filters are all too audible people have gone on a sampling rat race that still hasn't abated. DXD anyone? As a sanity check I would suggest that people try a seriously good sample rate conversion software like SaRaCon. Take a 96kHz file, take it down to 44.1kHz and then back to 96kHz to eliminate systematic errors. Now play both back and be ready to be amazed how transparent the process can be.
The upshot of all of this is: do not draw conclusions from pure tones, but on the other hand the 20kHz limit is pretty safe unless you deliberately push your luck.
Now, what does that mean for amplifiers?
1) Even if the audibility of >20kHz content may be subject to discussion, the fact that <20kHz content is much more audible than >20kHz content is not controversial anywhere. I've never heard it claimed that above 20kHz is more important than below, even by the most seriously deluded woo-woos.
That means that trying to optimize for supersonic performance is always a losing battle because even if 1dB of performance gained outside the audio band only incurs a 1dB performance loss inside the audio band, the performance loss is always much, much more readily audible than the supposed improvement. In fact, the noise shaping theorem says that the trade-off is much, much worse than a dB for a dB.
2) Limited amplifier bandwidth can only be a liability if the roll-off is very steep or the phase response is not linear. A Bessel (Thomson) filter has linear phase inside the bandwidth which equates only to a delay. Ncore has a mainly 1st order roll-off, which is Bessel by definition.
3) If you want to know something sonically meaningful about the high-frequency distortion of an amp, do a high-frequency IMD test. If the amplifier survives that, you can stop worrying about distortion above 20kHz.
Hi,
From Bruno: Even if the audibility of >20kHz content may be subject to discussion, the fact that <20kHz content is much more audible than >20kHz content is not controversial anywhere. I've never heard it claimed that above 20kHz is more important than below, even by the most seriously deluded woo-woos.
With respect, what kind of concept is this?
contains nothing. it is obvious that I have the first 20Khz.
You do not believe it makes sense to extend the bandwidth? or is it too complicated?
From Bruno: Even if the audibility of >20kHz content may be subject to discussion, the fact that <20kHz content is much more audible than >20kHz content is not controversial anywhere. I've never heard it claimed that above 20kHz is more important than below, even by the most seriously deluded woo-woos.
With respect, what kind of concept is this?
contains nothing. it is obvious that I have the first 20Khz.
You do not believe it makes sense to extend the bandwidth? or is it too complicated?
The question therefore becomes: is there a frequency where we can safely say that the ear has thrown in the towel? Only listening tests (I mean good ones, not acoustic wine tasting sessions) can answer this. Since it remains a hot topic in some circles, the tests keep coming and the results are consistent: you can insert a 20kHz low-pass filter undetectably provided it is not too steep. Amazingly the sort of filter we use in AD/DA is still gentle enough. What matters is that the filter is free from pre-echos and other nonidealities (which most AD/DA filters aren't
For that matter, transition bands are quite noticeable for lowpass filters in the audio band. The chief technologist at a previous job asked me to design and code up LPFs at 10 kHz with a 1 kHz transition, but they sounded raspy and unpleasant on music. I convinced him to try a 2 kHz transition, which sounded much less horrible. To forestall the inevitable question, the system ran well below clipping at all times.
Extremely steep anti-aliasing filters are the wrong way to go, IMO, precisely because of the pre-echo issue, but linear phase will always have some. On the other hand, one can design a minimum-phase FIR with a controlled rolloff which would show improved transient response at the cost of a bit more math.
If I quit the band to 20Khz, the integration does not move, it stays out of the band. I do not see complicated.
I followed the speeches and I realized that you had set the frequency response to 20Khz.
well, then we will all be looking measures at 60kHz.
@Francois Try not to confuse pre-ringing with pre-echo. Pre-ringing in filters above 20kHz is fairly innocuous (unless you go for fantastically sharp filters). Pre-echo always manifests itself over the whole bandwidth.
@AP2 The only thing that happens at 20kHz in Ncore or UcD is that that's the frequency where loop gain drops below the value at DC. You're being deliberately selective in your reading and I think from now on I'll be selective in what I bother replying to.
@AP2 The only thing that happens at 20kHz in Ncore or UcD is that that's the frequency where loop gain drops below the value at DC. You're being deliberately selective in your reading and I think from now on I'll be selective in what I bother replying to.
Last edited:
I did not ask you, as the band filter at 20Khz(what way have chosed). I do not want to know.
thanks for the "selective". certainly you have experience in dealing with people.
I might be missing something very obvious here, but isn't the amp performance well above 20k also very critical nowadays? With the modern noise shaping sources with more or less "gentle" output filters, the amp IMD products of out of band components will get folded into the audible range. So, unless there is a brick wall filter between the DAC and the amp or the amp IMD performance above 20k is spec-d, I don't see how one can ignore the elephant in the room.
Last edited:
You're not missing something. It is important that HF noise doesn't get demodulated. This is one reason why it's not a good idea to try to increase bandwidth too far beyond the power bandwidth.
In the case of linear amplifiers the problem is particularly tricky because demodulation happens outside the feedback loop (at the input stage) so the feedback loop is powerless against any in-band products that turn up.
In an amp like Ncore on the other hand (and many other class D amps) the only place where you can get demodulation is inside the feedback loop, which comes down like a ton of bricks on any audio-band products that might have otherwise resulted. So when I say that what matters is performance <20kHz I do mean: including any IMD products that might land there. So I should be more specific: performance for output signals up to about 20kHz (and automatically a goodish bit beyond that) but for input signals however high in frequency they come. So for instance, if harmonic distortion for a 50kHz signal is 0.1%, I don't care. If the same 50kHz signal causes in-band spuriae that is a problem. Thanks to control theory I can get good in-band audio performance without having to optimize supersonic performance. I have an old SACD player that puts out way more HF noise than a modern DAC. I can drive the amp flat out without any audible demodulation.
I think the way in which I might be confusing people is that it sounds like I think amps can have horrid performance above 20kHz. That's definitely not what I'm saying. What I'm saying is that I won't bend over backwards to get a 200kHz bandwidth if that adversely affects performance below 20kHz. Neither will I bend over backwards to make THD at 50kHz better if that results in worse performance at audible frequencies. But make no mistake about it: this amp is more than ordinarily capable at supersonic frequencies. Just not unnecessarily so.
In the case of linear amplifiers the problem is particularly tricky because demodulation happens outside the feedback loop (at the input stage) so the feedback loop is powerless against any in-band products that turn up.
In an amp like Ncore on the other hand (and many other class D amps) the only place where you can get demodulation is inside the feedback loop, which comes down like a ton of bricks on any audio-band products that might have otherwise resulted. So when I say that what matters is performance <20kHz I do mean: including any IMD products that might land there. So I should be more specific: performance for output signals up to about 20kHz (and automatically a goodish bit beyond that) but for input signals however high in frequency they come. So for instance, if harmonic distortion for a 50kHz signal is 0.1%, I don't care. If the same 50kHz signal causes in-band spuriae that is a problem. Thanks to control theory I can get good in-band audio performance without having to optimize supersonic performance. I have an old SACD player that puts out way more HF noise than a modern DAC. I can drive the amp flat out without any audible demodulation.
I think the way in which I might be confusing people is that it sounds like I think amps can have horrid performance above 20kHz. That's definitely not what I'm saying. What I'm saying is that I won't bend over backwards to get a 200kHz bandwidth if that adversely affects performance below 20kHz. Neither will I bend over backwards to make THD at 50kHz better if that results in worse performance at audible frequencies. But make no mistake about it: this amp is more than ordinarily capable at supersonic frequencies. Just not unnecessarily so.
Last edited:
I state that I'm talking to Bruno, because he is an engineer, and not becouse Hypex.
Your justification for limiting the bandwidth is absurd.
the problem of aliasing filter, if it is multi-pole, or if DAC is very good, or if the output buffer is guessed, let those who develop the DAC Professional.
it is obvious that if you want to 50Khz bandwidth and get real high-end performance.
IMD-thd-Linearity Low-latency-perfect clips and 104-db SNR that is really complex.
This can be a project that has value.
NuForce has a very wide band, amp has no noise (sounds dumb).
DXA have 60KHz-1dB, IMD perfect and noise-104dB.
if you present the measures extended to 60kHz or 50Khz, and known issues of class D are all resolved, then you have an amplifier of prestige.
But it is useless to say that more than 20Khz, because the noise is demodulated within the loop. then? this is just not easy to solve. are ten other things. Rest test of listen.. as others.
Just when extend the band and all performances remain good is goal!
Or because you are "Bruno", then I can not dare to ask? in Italy does not work type of name. works that demonstrate. (from Prof. up to the scientist).
Regards
Your justification for limiting the bandwidth is absurd.
the problem of aliasing filter, if it is multi-pole, or if DAC is very good, or if the output buffer is guessed, let those who develop the DAC Professional.
it is obvious that if you want to 50Khz bandwidth and get real high-end performance.
IMD-thd-Linearity Low-latency-perfect clips and 104-db SNR that is really complex.
This can be a project that has value.
NuForce has a very wide band, amp has no noise (sounds dumb).
DXA have 60KHz-1dB, IMD perfect and noise-104dB.
if you present the measures extended to 60kHz or 50Khz, and known issues of class D are all resolved, then you have an amplifier of prestige.
But it is useless to say that more than 20Khz, because the noise is demodulated within the loop. then? this is just not easy to solve. are ten other things. Rest test of listen.. as others.
Just when extend the band and all performances remain good is goal!
Or because you are "Bruno", then I can not dare to ask? in Italy does not work type of name. works that demonstrate. (from Prof. up to the scientist).
Regards
I see. I haven't run across this pre-echo so what came to mind was the pre-ringing familiar to anyone who's worked with linear phase FIRs. How does pre-echo manifest itself in a system? It implies a fair amount of stored state, so that's of some interest to me.
- Status
- Not open for further replies.