What makes a high end audio amp "better" then a low end unit?

[David Hess]

Thanks David, I hadn't come across the LT1166.

I just noticed that TI has discontinued the LME49830 and all similar drivers in favor of their class-D audio products but I do not really consider that a loss since a discrete implementation is straightforward.  The LT1166 has been around for quite some time now and there is no indication that it will be discontinued although I am not aware of it being used in commercial designs.

[Fungus]

Trust me, starting at 14KHz, I ca clearly hear it and each frequency up, it gets fainter.  I hear nothing at all on the 19 and 20 Khz.

So, about average then.

For 18Khz, I need to crank up the volume to the max on my headphones.  So, I can still detect 18k...

It's well below the -3dB point, though, right?

[BrianHG]

For 18Khz, I need to crank up the volume to the max on my headphones.  So, I can still detect 18k...

It's well below the -3dB point, though, right?

Yup.  I'm guessing even below -10db...
I figure within 10 years, my limit will be down to 16Khz, then, things will acceleratingly get worse...

As for me at 50 being average, this I'm not so sure.  Non of my friends in the past even at the age of 40 were still able to hear a CRT's 15.7KHz which I can still hear.  You do know about the fly 16KHz buzzing cell phone ring-tone which teenagers use in class since their teacher's even as young as 30 are not able to hear their phone ringing or receiving a msg...

[Fungus]

For hearing within first 20KHz, trying to reproduce the audio by sampling at 44.1k, it's the same effect of having a digital scope with a sampling rate of only 2x the max bandwidth.  With a repetitive signal, like most instruments, a quality up-sampling will do most of the time.  When it comes to metallic instruments which struck or hit,  the initial contact which might need to be a nasty squareish pulse only 1 to 4 sine cycles long will be messed up and it's location imaged by it's phase between the left and right channel might be missing the correct amplitude on one channel VS the other as you go beyond 15KHz.  This problem doesn't exist in analog formats or 96KHz recordings. 

Oh, dear.

Somebody hasn't watched this enough times:

[Fungus]

As for me at 50 being average, this I'm not so sure.

I'm over 50 and I can still hear something similar to what you claim, ie. 18kHz is definitely there when I turn the volume up, even if it's hard to say it's a "tone".

In musical terms though, I doubt frequencies that high are adding much to the performance, let alone the 70kHz you claim is on your vinyl (which I doubt is coming from any instrument, it will be an artifact of the recording).

You do know about the fly 16KHz buzzing cell phone ring-tone which teenagers use in class since their teacher's even as young as 30 are not able to hear their phone ringing or receiving a msg...

I heard about that. Even if we assume their phones can reproduce such a tone I still think it's a urban myth, along with ultrasonic pest repellents.

[paulca]

I have a query which relates to this frequency discussion, though I am coming from the digital side.

Sample Theorem and Nyquist Theorem suggests that to capture a signal you must have 2*f bandwidth.  Now this makes sense in one level, in that for each wave you need at least a high and a low.

However, a 10Khz sine wave cannot be represented accurately within a 20khz bandwidth.  At 20Khz the 10KHz sine wave will turn into a 10Khz saw tooth wave, and that's assuming it's correctly phase aligned.  To fairly sample a sine wave of frequency f you would need much more than 2*f bandwidth.  I realise that filters will smooth the saw wave, but that's not a perfect reflection of the original.

Maybe I'm thinking far too digital sample theorem here and in analogue signals it doesn't matter as much.

I noticed oscilloscopes recommend 5*f minimum.  is this so that you at least get a 5 sample reflection of the original waveform shape?

[Fungus]

However, a 10Khz sine wave cannot be represented accurately within a 20khz bandwidth.  At 20Khz the 10KHz sine wave will turn into a 10Khz saw tooth wave

You need to watch the video I posted above a few times.

Signal reconstruction is not "join the dots with straight lines".

that's assuming it's correctly phase aligned.  To fairly sample a sine wave of frequency f you would need much more than 2*f bandwidth.

You're half right. You need more than 2*f, yes, but not very much more (in theory only a tiny fraction more, in practice a bit more than that... up to 2.5*f is common on things like oscilloscopes).

[AndyC_772]

With a 10 kHz sine wave sampled at 20 kHz, you could be unlucky and sample precisely at each zero crossing, so the output would be all zeros. The thing to note here is simply that you need a sample rate which is more than twice the highest signal frequency, not just exactly twice the highest frequency.

What is often misunderstood is that there is only one correct way to reconstruct an analogue signal from digital samples, and it's not joining the dots with straight lines.

Mathematically, there is only one solution to the problem of fitting a curve through the sampled points which does not include frequency components above 0.5x the sampling rate. Two common incorrect methods - linearly interpolating between samples with straight lines, and holding a fixed output between one sample time and the next - both introduce higher frequency components which do not exist in the original data.

It's interesting to note also that the mathematically correct solution yields precisely the same waveform regardless of the relative timing between the original signal and the sampled points. A common misconception is that if a portion of a signal is sampled near its peaks, then the reconstructed portion will have greater amplitude than if it's sampled away from them. This is just plain wrong, unless the discussion is specifically about the practicalities of different reconstruction filters.

Faster sampling may in many cases may yield a better result than sampling at the bare minimum required to meet the Nyquist criterion, but the reason is to do with practical limitations of real filters.

[Fungus]

What is often misunderstood is that there is only one correct way to reconstruct an analogue signal from digital samples, and it's not joining the dots with straight lines.

Monty's video (posted above) has the best/clearest explanation I know of.

(It can take several viewings to fully grok though ... at least in my case)

[paulca]

Mathematically, there is only one solution to the problem of fitting a curve through the sampled points which does not include frequency components above 0.5x the sampling rate.

I did this all at Uni in Digital Communications, but it's all leaked out, probably into beer cans at some point.

I do remember there was something about an infinite number of solutions to the sample caused by octave harmonics and some such.  Digital samples have aggressive cut off filters to remove these harmonics.  For example if you sample a 20Khz signal at 10Khz you would appear to have a 10Khz waveform at the end, when of course you don't, you have just sampled every other wave.  Apparently it goes the other way too.

Anyway, something like that, as I said most of it has leaked out from between my ears as I work purely in software these days I don't really have to deal with ADC DACs.

I've added the video above to my list to watch later.

[coppice]

Signal reconstruction is not "join the dots with straight lines".

To cover both the common misconceptions about sample reconstruction, reconstruction is not:

• Joining the dots with straight lines, or
• Drawing steps between the samples, so the waveform resembles a histogram

The samples are a series of instantaneous points on the original waveform. There is only one waveform which is band limited to half the sample rate which will precisely join all of these points. So, you apply the stream of instantaneous samples to a suitable band limiting filter, and out comes precisely the original waveform.

Once you understand what sampling and reconstruction is really about, the join the dots idea just looks wacky. However, the stepped waveform idea has important merit in the real world. Its approximately what you get out of most DACs. Look up "zero order hold" and you will find a full analysis of this kind of system. Once you band limit the stepped waveform to half the sample rate you end up with something similar to the original signal, but with its high frequency response rolled off in a very predictable way. To precisely recreate the original signal you need to compensate for this roll-off by either boosting the highest frequencies in the band with a digital filter before the DAC, or use a similarly non-flat analogue filter, in addition to the band limiting filter, after the DAC.

One caveat in this talk of signal filtering is you need to be careful about the phase response of any filters used, if you want to recover the original waveform precisely.

[Fungus]

For example if you sample a 20Khz signal at 10Khz you would appear to have a 10Khz waveform at the end, when of course you don't, you have just sampled every other wave.

Obviously true, but the stuff they put on CDs (for example) is bandwidth limited (ie. filtered) to remove everything above the Nyquist limit before they sample it.

Everything above the 22kHz Nyquist limit of CDs is inaudible anyway. It adds nothing to the sound.

Bottom line: 16bit, 44.1kHz is enough to perfectly reproduce audio, despite what the "golden ears" might tell tell you.

In theory a case could be made for 48kHz sampling because it allows the low pass filter applied at the mastering stage to be a bit less aggressive. In practice it doesn't seem to make much difference. 96 or 192kHz? Not needed. 24 bits? Not needed.

It's all in the video.

[Loboscope]

Be aware, that anything in human audio is bandwith-limited. So are our ears, even if some members here seem to be anatomic miracles, at least exceptions  . For further information please look here: https://www.amazon.com/Auditory-Neuroscience-Making-Sense-Sound/dp/0262518023.
But all natural sources of sound are bandwith-limited too. Because of the mass-inertia of strings, reeds, resonance-wood, drum-membranes, speaker-membranes a.s.o., it will take some times for the transients to rise. Therefore audio will never deal with square-waves, the rise-times of human audio signals will be limited by nature. Even the compressibility of the air itself will prohibit square-waves being transferred.
Additionally all high-frequency content of Instruments/Voices are of low level in relation to the fundamental frequencies.
So no ADC / DAC in audio-systems has to deal with square-waves and/or a 20 KHz-wave at 0 dB for example.
A tiny bit more than 2*f as Nyquist-frequency will be enough to sample all of our wonderful music we here all days, 44,1 KHz will be enough. And as distribution format the CD with 16 bit / 44.1 KHz will be sufficient.

The audio quality depends only from the quality of the sources (first and most of all the musicians!), the recording an the responsible recording engineer and his mixing/mastering qualities.
I have lots of excellent CD´s here and I make recordings as sideline profession. My main profession is musician with classic-academic studies and I know how instruments, how music will sound.
After I bought my first CD-Player in the late eighties last century I no more touched my Vinyl again.

[coppice]

Bottom line: 16bit, 44.1kHz is enough to perfectly reproduce audio, despite what the "golden ears" might tell tell you.

16 bits properly dithered is fine for distribution. Without dithering its a bit iffy. Its a pity they didn't go for something like 17 or 18 bits, and put the adequacy of the dynamic range beyond question. People seem to get hooked on round numbers like 16 and 24.  . The bottom line is 16 bits was not great in the early 80s, when 16 bit/44.1ksps was introduced, but as our understanding of dithering techniques has developed, it has become adequate.

44.1ksps also seems to be perfectly adequate. Being able to hear a tone at 23kHz or 24kHz with reasonable sensitivity doesn't mean that content at those frequencies will have any material effect on your perception of complex musical sounds. In my youth, when I could hear those frequencies pretty well, I was never aware of the lack of them affecting my perception of music. A nice flat response to 20kHz seems fine. Its actually at the other end of the spectrum where systems are usually lacking. Typically 20Hz is considered the bottom of the range of human hearing, However, the lowest note on the pedals of a large organ is in the 13 to 16Hz range (depending on the organ). You feel more than hear those notes, but they are a very material part of the musical experience in a cathedral. Even today, when most of the technical reasons for poor bass response have been dealt with, few audio system reproduce that stuff well. A honk at the resonance of the speaker is all most systems seem to muster.

[BrianHG]

For hearing within first 20KHz, trying to reproduce the audio by sampling at 44.1k, it's the same effect of having a digital scope with a sampling rate of only 2x the max bandwidth.  With a repetitive signal, like most instruments, a quality up-sampling will do most of the time.  When it comes to metallic instruments which struck or hit,  the initial contact which might need to be a nasty squareish pulse only 1 to 4 sine cycles long will be messed up and it's location imaged by it's phase between the left and right channel might be missing the correct amplitude on one channel VS the other as you go beyond 15KHz.  This problem doesn't exist in analog formats or 96KHz recordings. 

Oh, dear.

Somebody hasn't watched this enough times:

You need to watch the video I posted above a few times.

Signal reconstruction is not "join the dots with straight lines".

He is not completely right, unless the audio output equipment has proper analog filtering, or sin(x)/x correction.  However, since sound samples in a PC rarely match the DAC's output sample rate today, this filtering must also be done in Windows or in the sound card drivers.  Do you trust Microsoft, or all the apps you installed on you PC.
Look here, I measured these an hour ago, see how it looks for real on my scope: (I matched his setup, but specified my sample rates...)

#1, 20Khz sine wave sampled and played at 96KHz:

#2, 20Khz sine wave sampled and played at 44.1KHz:

#3, 20Khz sine wave sampled and played at 44.1KHz, however, I used a DSP feature built into Foobar2000 music player called PPHS resampler set to 96KHZ, ultra mode:


Lesson, your video is only correct if proper analog filtering practice is applied, which I already knew and also knew about how today, only a few % of audio equipment bother to attend to this issue properly accommodating multiple source sample rates.  Lesson 2, if you have a PC as your music player with a 96KHz DAC, and you want to bypass the system re sampling to guarantee a good output sine wave at the high above 16KHz when playing CD 44.1KHz music, use Foobar2000 player with the PPHS resampler enabled and set to 96K, ultra mode and your analog output will be guaranteed have a properly clean sin(x)/x filtered 20Khz signal reconstruction.

In other words, if and only if you playback hardware has properly tuned analog output filtering, or may be done in combination with digitally filtered upsampling and analog filtering for that new sample rate, under these circumstances, a stereo 44.1KHz sample should achieve at least a 22Khz bandwidth reproduction with proper stereo image.  For those who try to nit-pick specific lemon multiple tones, I'm sorry to say that no human instrument performance would ever sit exactly on a lemon frequency exactly.  For those who sample and play at 96KHz, you are just pretty much guaranteeing that errors in sampling and playback hardware wont produce the nasty problem in my scope shot #2 for frequencies between around 16KHz and 20KHz.  Errors now at the new higher frequencies due to poor design practices are way too far outside our hearing to bother with and microphones in the recording studio have already lost soo much db at 40Khz anyways that there isn't much signal left.

Source audio files available upon request.
Freeware Foobar2000 media player with SDK: http://www.foobar2000.org/

[BrianHG]

I know how Monty Montgomery @ xiph.org says true that the sample data represent a smooth linear signal and properly designed AD/DA systems should reconstruct the original smooth analog wave of the source up to 1/2 the sample frequency.  But, unfortunately, today, consumer grade equipment don't quite measure up in reconstructing the original source signal's true image.  It's funny how older non-oversampling DACs in really old CD players with good analog filters at the output stage may do this reconstruction better than some of today's technology when it comes to dealing specifically with 44.1k 16 bit audio.

[hamster_nz]

@BrianHG, what happens if you play the 44.1 sample, and turn on the channel's low pass filter, set to 22kHz (like in a good reconstruction filter)?

If you then flip it to high pass you should see the 24.1kHz alias in there...

[AndyC_772]

I do remember there was something about an infinite number of solutions to the sample caused by octave harmonics and some such.

I think you're just describing aliasing. If you have a component in the original analogue signal at a frequency exceeding 0.5*fs, then it'll appear in the reconstructed signal at a lower frequency, and is impossible to distinguish from a real, wanted signal at that same lower frequency.

That's why a sampling system has to include an analogue low-pass filter in front of the ADC.

[BrianHG]

@BrianHG, what happens if you play the 44.1 sample, and turn on the channel's low pass filter, set to 22kHz (like in a good reconstruction filter)?

If you then flip it to high pass you should see the 24.1kHz alias in there...

I chose 20KHz since it had really pronounced garbage on the scope.  I assume generating a 22.05KHz sine would show perfectly.
As for Foobar2000's PPHS resampler, is has no option other than to select an output sample rate and 'ultra' mode.  I assume when I send it a 44.1k source audio, it is precisely tuned to properly limit filter that to 22.05khz.

As for your idea, I grabbed a frozen scope shot of my #2 illustration:  However, this is not the same as over shooting the source frequency bandwidth.  Thus again, sampling at 96k allows for a super refined low pass filter to get rid of anything above 22.05khz when downsampling to 44.1k.  Again, some programs do a better job than others when doing this but I don't trust Window's built in resampling.  I would always run my audio acquisition software at the specified rating of it's ADC, the do a thorough downsample to my operating frequency.  With hard drive space today, I'm happy to just stick with 96k.  192k is just a waste...

[Fungus]

I know how Monty Montgomery @ xiph.org says true that the sample data represent a smooth linear signal and properly designed AD/DA systems should reconstruct the original smooth analog wave of the source up to 1/2 the sample frequency.  But, unfortunately, today, consumer grade equipment don't quite measure up in reconstructing the original source signal's true image.

Weird how Monty's 15 year old consumer grade sound card can do it perfectly...

[BrianHG]

I know how Monty Montgomery @ xiph.org says true that the sample data represent a smooth linear signal and properly designed AD/DA systems should reconstruct the original smooth analog wave of the source up to 1/2 the sample frequency.  But, unfortunately, today, consumer grade equipment don't quite measure up in reconstructing the original source signal's true image.

Weird how Monty's 15 year old consumer grade sound card can do it perfectly...

15 years ago, he has a 192Khz capable sound card, back then, that's not consumer grade.  That's a studio grade sound card.  (You can see he is able to select higher sample rate settings....)
The other possible side of the coin goes to my other statement above:  Older hardware specifically tuned to 44.1k only has hard wired analog output filters specifically limited to 22.05KHz.  Modern engineers have gotten sloppy with DACs which have built in voltage out driving the line out/headphones directly without an analog op-amp and sin(x)/x filter.

Even my old Soundblaster AWE64Gold from 98 had op-amp I-V converter circuit with a filter stage.  I know because I modified the gain resistor to get an excellent 5vp-p output with almost no noise at all.  (Great for DVD movies at the time with very low volume recordings, or wide dynamic ranges...)

[Zero999]

Bottom line: 16bit, 44.1kHz is enough to perfectly reproduce audio, despite what the "golden ears" might tell tell you.

16 bits properly dithered is fine for distribution. Without dithering its a bit iffy. Its a pity they didn't go for something like 17 or 18 bits, and put the adequacy of the dynamic range beyond question. People seem to get hooked on round numbers like 16 and 24.  . The bottom line is 16 bits was not great in the early 80s, when 16 bit/44.1ksps was introduced, but as our understanding of dithering techniques has developed, it has become adequate.

I agree with you that more than 16-bits is required for recording purposes, because it avoids the need to keep adjusting the gain control or a limiting circuit, so the normal level can be set so it never clips. However, for playback, I think 16-bit is more than adequate, because music doesn't use the full  undithered 90.3dB dynamic range anyway.

44.1ksps also seems to be perfectly adequate. Being able to hear a tone at 23kHz or 24kHz with reasonable sensitivity doesn't mean that content at those frequencies will have any material effect on your perception of complex musical sounds. In my youth, when I could hear those frequencies pretty well, I was never aware of the lack of them affecting my perception of music. A nice flat response to 20kHz seems fine.

I agree. Those faint high pitch sounds, will be drowned out by the lower frequency content. I also believe pitch perception declines at higher frequencies, so it's less important.

How about recording at much higher sample rates? It makes more sense to avoid rounding errors when doing intermediate processing but is it needed for recording? I suppose the master should be as perfect as possible and not having the lag and phase distortion of a filter could make it easier.

Its actually at the other end of the spectrum where systems are usually lacking. Typically 20Hz is considered the bottom of the range of human hearing, However, the lowest note on the pedals of a large organ is in the 13 to 16Hz range (depending on the organ). You feel more than hear those notes, but they are a very material part of the musical experience in a cathedral. Even today, when most of the technical reasons for poor bass response have been dealt with, few audio system reproduce that stuff well. A honk at the resonance of the speaker is all most systems seem to muster.

It's possible that many people can't hear the fundamental of a <16Hz organ note, only the harmonics. The lower frequencies aren't important for pitch perception at the bass end of the spectrum. The fundamental can be missing, yet the brain will still perceive the correct pitch.
https://en.wikipedia.org/wiki/Missing_fundamental

The ears and the brain can be easily fooled, no matter how cleaver one is. The key is to beware of this, especially when someone tries to part you from your hard earned money.

[BrianHG]

It's possible that many people can't hear the fundamental of a <16Hz organ note, only the harmonics. The lower frequencies aren't important for pitch perception at the bass end of the spectrum. The fundamental can be missing, yet the brain will still perceive the correct pitch.
https://en.wikipedia.org/wiki/Missing_fundamental

The ears and the brain can be easily fooled, no matter how cleaver one is. The key is to beware of this, especially when someone tries to part you from your hard earned money.

Now I'm not saying you need slightly better quality than a CD, I'm saying with today's technology, you should be able to get at least slightly better quality than CD quality for free, at no extra cost.  Anyone going off the deep end trying to sell you a 1MHZ amp, picovolt accuracy, silver plated copper speaker cable are full of audiophoolery BS.  I just expect that at a minimum, you shouldn't see my second scope shot illustration #2 from my laptop, just as fungus said, a 15 year old sound card didn't exhibit that modulation problem with a 20KHz tone and I could only get rid of the problem with special dedicated software.  So much for a laptop with a 192KHz, 24 bit audio.  It plays the Hi-def fine, but, cant even cope with CD 44.1k 16bit without special software countermeasures.

[Loboscope]

It's possible that many people can't hear the fundamental of a <16Hz organ note, only the harmonics. The lower frequencies aren't important for pitch perception at the bass end of the spectrum. The fundamental can be missing, yet the brain will still perceive the correct pitch.
https://en.wikipedia.org/wiki/Missing_fundamental

The ears and the brain can be easily fooled, no matter how cleaver one is. The key is to beware of this, especially when someone tries to part you from your hard earned money.

16 Hz will only be reproduced by organs which have a 32´-stop. There are not so much organs with 32´, only the greatest ones have these deep going stops. I guess 99% of all organs in the world will only have 16´-stops as deepest stop. Will say, that they ´only´ will reproduce 32 Hz as deepest tone.
There are stops which use the phenomenon of the "missing fundamental" by creating deep notes with two pipes sounding together at a distance of a fifth (Quinte). With the proper intonation, this lets sound the sub-octave of the deeper tone of the pipes, a so called combination-tone (Kombinationston). The level of this sub-octave will be lower than the level of of the original tones of the pipes, but our ear/brain still recognizes/reconstructs the deeper tone.
Also the fundamentals of great bells have lower levels than their harmonics, but here also our ear/brain recognizes the fundamental and the bell will sound deep and sonorous.


Concerning the capabilities of modern ADC/DAC-converters I did run a simple test on my Behringer X32-Rack. Behringer is known as a brand that stands for less expensive but normally good constructed gear and equipment for PA and recording intended for semi-professional and professional use. (For the X32 family and the digital mixer lines look here: http://www.music-group.com/Categories/Behringer/Mixers/Digital-Mixers/c/1234111?group=Digital%20Mixers&colExpFlag=,Digital%20Mixers)

I did run a 20, 21 and 22 KHz test signal through the mixer analog in and analog out, so it has to pass the ADC and the DAC (at a samling rate of 44,1 KHz). The level was a tiny bit below 0 dB, short before clipping. These are signals, the converters has not to deal with in real life, because only very few instruments will have harmonics up to 20 KHz or even more, and if, their level will be very, very low.

Below, there are the three samples as scope-snapshot. You will see, that 20 KHz is fine, 21. KHz is a little bit distorted and 22 KHz (at ca. -0,5 dB at the X32!) will be distorted obviously by an artifact.
But as I said, this will never disturb any audio-source coming from a natural instrument/sound.

As far as I know, Behringer uses ADC´s/DAC´s from AKM and certainly not the most expensive ones. If you by ultra-cheap crap, you will run into problems, don´t matter if it is an amp, a CD-Player, a speaker a.s.o.
But any gear from the middle-range upwards shall guarantee a good to very good sound quality. My nearfield Monitor-Speakers are Neumann KH120A [http://www.neumann-kh-line.com/neumann-kh/home_en.nsf/root/prof-monitoring_studio-monitors_nearfield-monitors_KH120A] and they sound precise, clean and neutral as a Monitor for mixing/mastering should do. They are not cheap, but also not expensive, mid-range in price I would call them. So nowadays you will not have to invest a fortune to have excellent audio equipment.

[BrianHG]

As far as I know, Behringer uses ADC´s/DAC´s from AKM and certainly not the most expensive ones.

Even AKM's entry level DACs are really good and properly designed.  If I remember correctly, they have their own proprietary I/V technology stage built into their audio voltage output DACs.