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

[Fungus]

As you stated, most, if not all modern audio DACs are sigma delta, - their output is updated at 64/128x the sampling rate, pushing the sample rate waaay up the spectrum, meaning the a simple output filter, combined with the frequency response of an amp greatly attenuates it. 

Yep, I've done that and it works.

Ideally you should output curves, not straight lines though.

[coppice]

As you stated, most, if not all modern audio DACs are sigma delta, - their output is updated at 64/128x the sampling rate, pushing the sample rate waaay up the spectrum, meaning the a simple output filter, combined with the frequency response of an amp greatly attenuates it.  I won't go into interpolation/digital filters etc.. as ultimately it just means that at the output, the sample frequency is now in the MHz range, and is already attenuated.  Unless one is plugging the output of their DAC into an RF amp, I don't see why output filtering is particularly important. Yes, it is needed, but I can't see it being something that requires a great deal of design?

I didn't say it was a tough job to do the filtering these days. I said it was needed in response to someone saying it was unnecessary. A huge number of audio amps really do not like RF, and are poorly protected from its effects. Everybody knows the burbling pattern of GSM interactions because of this.

Note the last one isn't snake oil or anything, I'm sure its a very well designed R2R DAC, but seems like a LOT of effort and money for something that performs very similarly to a $2 IC.

Good stereo DACs haven't cost $2 for a very long time. 20 cents is more like the going rate. Its crazy these days. A good op-amp to buffer the output of the DAC can cost more than the DAC.

[David Hess]

The output of the DAC still needs to be filtered because intermodulation from non-linearity in the output amplifier which increases with frequency will result in mixing products in the audible band and a general increase in noise.

The same thing happens with speakers when a single driver handles a broad range of frequencies and has a large displacement.  This is especially a problem with woofers at low frequencies where displacement may be large and the intermodulation produces a "boomy" sound.  Base reflex and horn designs help prevent this by lowering mechanical impedance to produce more power with less cone movement.

[Zero999]

As you stated, most, if not all modern audio DACs are sigma delta, - their output is updated at 64/128x the sampling rate, pushing the sample rate waaay up the spectrum, meaning the a simple output filter, combined with the frequency response of an amp greatly attenuates it.  I won't go into interpolation/digital filters etc.. as ultimately it just means that at the output, the sample frequency is now in the MHz range, and is already attenuated.  Unless one is plugging the output of their DAC into an RF amp, I don't see why output filtering is particularly important. Yes, it is needed, but I can't see it being something that requires a great deal of design?

I didn't say it was a tough job to do the filtering these days. I said it was needed in response to someone saying it was unnecessary. A huge number of audio amps really do not like RF, and are poorly protected from its effects. Everybody knows the burbling pattern of GSM interactions because of this.

Yes cheap amplifiers often overlook filtering and good EMC design practise.

The output of the DAC still needs to be filtered because intermodulation from non-linearity in the output amplifier which increases with frequency will result in mixing products in the audible band and a general increase in noise.

A high end amplifier shouldn't do that, because it will behave at 44.1kHz and have good filtering for the harmonics which will cause problems.

[David Hess]

The output of the DAC still needs to be filtered because intermodulation from non-linearity in the output amplifier which increases with frequency will result in mixing products in the audible band and a general increase in noise.

A high end amplifier shouldn't do that, because it will behave at 44.1kHz and have good filtering for the harmonics which will cause problems.

A high end amplifier has the same problem filtering the aliasing products outside of a 20 kHz bandwidth sampled at 44.1 kHz as the DAC reconstruction filter does.  Oversampling DACs make this into a trivial issue where a simple low order filter is sufficient but in the past, 11th order filters (1) were used to produce enough rolloff between 20 and 22 kHz.

The amplifier does not need to misbehave at higher frequencies to cause problems.  Non-linearity simply increases at higher frequencies producing mixing products which may end up in the audio band.

(1) At least that is what Burr-Brown required for their DACs to meet their specifications.

[Zero999]

The amplifier does not need to misbehave at higher frequencies to cause problems.  Non-linearity simply increases at higher frequencies producing mixing products which may end up in the audio band.

That's pretty much what I meant by misbehaving and is probably one of the reasons why many audio amplifiers are designed to work well above the audio band, so ultrasonic noise from CDs and other digital sources don't cause buzzing. For example: the LM3886 is specified for low distortion up to 20kHz and has a full power bandwidth of 80kHz and the NTE1380 has a FPBW of 140kHz. Somehow I doubt they're going to produce audible noise from 44.kHz.

http://www.ti.com/lit/ds/symlink/lm3886.pdf
http://www.nteinc.com/specs/1300to1399/pdf/nte1380.pdf

It's true a good filter will be required to block the higher 44.1kHz harmonics, which could do something funky, but I doubt the fundamental needs to be attenuated to ridiculously low levels, to avoid noise.

[paulca]

The high end amp costs more.  That's about all you know for sure.

[David Hess]

The amplifier does not need to misbehave at higher frequencies to cause problems.  Non-linearity simply increases at higher frequencies producing mixing products which may end up in the audio band.

That's pretty much what I meant by misbehaving and is probably one of the reasons why many audio amplifiers are designed to work well above the audio band, so ultrasonic noise from CDs and other digital sources don't cause buzzing. For example: the LM3886 is specified for low distortion up to 20kHz and has a full power bandwidth of 80kHz and the NTE1380 has a FPBW of 140kHz. Somehow I doubt they're going to produce audible noise from 44.kHz.

FPBW (full power bandwidth) is just another name for large signal response and depends on slew rate and output voltage.  The gain-bandwidth product or small signal bandwidth is more important for reducing distortion unless the full power bandwidth/large signal response/slew rate is insufficient which should never happen.

The NTE1380 datasheet does not show a schematic but the LM3886 does.  See those 1.1k resistors used for emitter degeneration of the input PNP differential pair?  They lower the input stage transconductance while maintaining the same tail current.  This allows the compensation capacitor marked as 10 picofarads to be smaller then it otherwise would need to be improving the output slew rate and full power or large signal bandwidth.  As a side effect, those 1.1k resistors in series with the emitters increase the input noise considerably.  If a schematic is not available or is incomplete, then the input noise specification can be used to determine if emitter/source degeneration was used.  Contrast the 318 and the 833; the former has 10 times the slew rate (and full power bandwidth) at the cost of 3 times higher voltage noise do to emitter degeneration to reduce transconductance.

The old 318 operational amplifier used emitter degeneration resistors to get its high frequency performance and was noisy and lower precision as a result.  Most audio power amplifiers do as well.  FET input amplifiers have lower transconductance to start with.  There are also other ways to handle transconductance reduction.

For audio power amplifiers, input noise is not normally a concern because it is assumed that a high level signal is available but I have noticed that many modern amplifiers produce a really annoying hiss with no input signal when connected to high efficiency speakers.

The excess gain at frequencies above the audio band is just a side effect of maximizing the gain at high frequencies to limit distortion.  I do not think anybody does it to limit intermodulation of content above the audio band which should not exist anyway and would not have been a problem before digitally sourced music.  But I do wonder if early inconsistent reports of digital audio quality came about because of inadequate anti-alias filtering allowing high frequency alias products to mix in the power amplifier.  The anti-aliasing filters were an obvious place to save money and lots of cheap audio gear did so.

It's true a good filter will be required to block the higher 44.1kHz harmonics, which could do something funky, but I doubt the fundamental needs to be attenuated to ridiculously low levels, to avoid noise.

Burr-Brown and others were not recommending 11th order anti-alias filtering with a transition band between 20 and 22 kHz to make their 44.1 kHz ADCs and DACs easier to sell.  This was such a big problem that oversampling ADCs and DACs with their much simpler anti-aliasing requirements replaced them quickly and made this a solved problem even on the economic side.

[Zero999]

The amplifier does not need to misbehave at higher frequencies to cause problems.  Non-linearity simply increases at higher frequencies producing mixing products which may end up in the audio band.

That's pretty much what I meant by misbehaving and is probably one of the reasons why many audio amplifiers are designed to work well above the audio band, so ultrasonic noise from CDs and other digital sources don't cause buzzing. For example: the LM3886 is specified for low distortion up to 20kHz and has a full power bandwidth of 80kHz and the NTE1380 has a FPBW of 140kHz. Somehow I doubt they're going to produce audible noise from 44.kHz.

FPBW (full power bandwidth) is just another name for large signal response and depends on slew rate and output voltage.  The gain-bandwidth product or small signal bandwidth is more important for reducing distortion unless the full power bandwidth/large signal response/slew rate is insufficient which should never happen.

The NTE1380 datasheet does not show a schematic but the LM3886 does.  See those 1.1k resistors used for emitter degeneration of the input PNP differential pair?  They lower the input stage transconductance while maintaining the same tail current.  This allows the compensation capacitor marked as 10 picofarads to be smaller then it otherwise would need to be improving the output slew rate and full power or large signal bandwidth.  As a side effect, those 1.1k resistors in series with the emitters increase the input noise considerably.  If a schematic is not available or is incomplete, then the input noise specification can be used to determine if emitter/source degeneration was used.  Contrast the 318 and the 833; the former has 10 times the slew rate (and full power bandwidth) at the cost of 3 times higher voltage noise do to emitter degeneration to reduce transconductance.

The old 318 operational amplifier used emitter degeneration resistors to get its high frequency performance and was noisy and lower precision as a result.  Most audio power amplifiers do as well.  FET input amplifiers have lower transconductance to start with.  There are also other ways to handle transconductance reduction.

For audio power amplifiers, input noise is not normally a concern because it is assumed that a high level signal is available but I have noticed that many modern amplifiers produce a really annoying hiss with no input signal when connected to high efficiency speakers.

The excess gain at frequencies above the audio band is just a side effect of maximizing the gain at high frequencies to limit distortion.  I do not think anybody does it to limit intermodulation of content above the audio band which should not exist anyway and would not have been a problem before digitally sourced music.  But I do wonder if early inconsistent reports of digital audio quality came about because of inadequate anti-alias filtering allowing high frequency alias products to mix in the power amplifier.  The anti-aliasing filters were an obvious place to save money and lots of cheap audio gear did so.

I agree about power audio amplifiers being a bit more noisy than op-amps. The NTE1380 is same as the old TDA2030, which I've tested and the noise was not detectable, even with headphones, although the gain was only 16, so that's hardly surprising: the spec is typically 3µV, over the audio bandwidth, which would give 48µV out.

EDIT:
I've just calculated the noise for the TDA2030/NTE1380 and it comes to 20.2nV/Hz^0.5 typical, 7.5nV/Hz^0.5. The current noise isn't too bad though: 1.35pA/Hz^0.5 If you need significantly more than the minimum allowed gain, use a pre-amplifier.

http://www.changpuak.ch/electronics/datasheets/TDA2030.pdf

It's true a good filter will be required to block the higher 44.1kHz harmonics, which could do something funky, but I doubt the fundamental needs to be attenuated to ridiculously low levels, to avoid noise.

Burr-Brown and others were not recommending 11th order anti-alias filtering with a transition band between 20 and 22 kHz to make their 44.1 kHz ADCs and DACs easier to sell.  This was such a big problem that oversampling ADCs and DACs with their much simpler anti-aliasing requirements replaced them quickly and made this a solved problem even on the economic side.

Sounds overkill. There are other sources of ultrasonic noise than digital audio: switch mode power supplies and RF, especially the IF in a receiver, which would have been a potential issue, long before digital audio was widespread. I couldn't get the old TDA2030 to produce any audible noise, when the input frequency was swept from 22kHz to 200kHz. The only problem I had was the zobel network resistor overheating, when I had the output level too high. I suppose that's a constant signal though, so that's not a fair test. I should've done it with an FM and AM modulated 1kHz tone.

[paulca]

Burr-Brown and others were not recommending 11th order anti-alias filtering with a transition band between 20 and 22 kHz to make their 44.1 kHz ADCs and DACs easier to sell.  This was such a big problem that oversampling ADCs and DACs with their much simpler anti-aliasing requirements replaced them quickly and made this a solved problem even on the economic side.

Slightly related, but I knew I'd see this name somewhere before.

My digital USB headphone amp identifies as a Burr-Brown.  I googled them, they appear to be a US audio IC company.

However, the USB ident information goes on to say "Burr-Brown Japan"

... and the actual DAC chip idents are "Texas Instruments PCM2702 16 bit stereo audio DAC"

Out of interest this was purchased as a "Pro-Ject Head Box" headphone amp.
https://www.amazon.co.uk/Pro-Ject-Head-Headphone-Amplifier-Black/dp/B00AMKZ7C4

Anyway, it's gorgeous, fat, clean, quiet and clear.  I love it.

[Buriedcode]

Burr Brown was aquired by TI in .. 2000 ?  I'm guessing they just kept the VID/PID

[David Hess]

I agree about power audio amplifiers being a bit more noisy than op-amps. The NTE1380 is same as the old TDA2030, which I've tested and the noise was not detectable, even with headphones, although the gain was only 16, so that's hardly surprising: the spec is typically 3µV, over the audio bandwidth, which would give 48µV out.

I was just pointing out why they are noisier and they are not noisier than all operational amplifier; the same limitations apply to both.  Emitter degeneration is the easiest way to increase the full power bandwidth and the added noise is not a problem with low efficiency speakers intended for solid state amplifiers.  But a solid state amplifier like this used with high efficiency speakers intended for a lower power vacuum tube amplifier results in an annoyingly loud hiss.

Burr-Brown and others were not recommending 11th order anti-alias filtering with a transition band between 20 and 22 kHz to make their 44.1 kHz ADCs and DACs easier to sell.  This was such a big problem that oversampling ADCs and DACs with their much simpler anti-aliasing requirements replaced them quickly and made this a solved problem even on the economic side.

Sounds overkill. There are other sources of ultrasonic noise than digital audio: switch mode power supplies and RF, especially the IF in a receiver, which would have been a potential issue, long before digital audio was widespread. I couldn't get the old TDA2030 to produce any audible noise, when the input frequency was swept from 22kHz to 200kHz. The only problem I had was the zobel network resistor overheating, when I had the output level too high. I suppose that's a constant signal though, so that's not a fair test. I should've done it with an FM and AM modulated 1kHz tone.

It *was* overkill for many manufacturers and they either left the anti-aliasing filters out or used cheaper lower order filters.  Besides the other causes like using the wrong equalization for mastering, I speculate that this was the cause of early "harsh" sounding CD audio.

An intermodulation test would be required.  With such a filter installed, there would be no test signal so no intermodulation.  There is little reason to test for this in an audio power amplifier; to borrow an aphorism from programming, do not trap errors that you cannot handle.

Burr-Brown and others were not recommending 11th order anti-alias filtering with a transition band between 20 and 22 kHz to make their 44.1 kHz ADCs and DACs easier to sell.  This was such a big problem that oversampling ADCs and DACs with their much simpler anti-aliasing requirements replaced them quickly and made this a solved problem even on the economic side.

Slightly related, but I knew I'd see this name somewhere before.

My digital USB headphone amp identifies as a Burr-Brown.  I googled them, they appear to be a US audio IC company.

However, the USB ident information goes on to say "Burr-Brown Japan"

... and the actual DAC chip idents are "Texas Instruments PCM2702 16 bit stereo audio DAC"

Out of interest this was purchased as a "Pro-Ject Head Box" headphone amp.
https://www.amazon.co.uk/Pro-Ject-Head-Headphone-Amplifier-Black/dp/B00AMKZ7C4

Anyway, it's gorgeous, fat, clean, quiet and clear.  I love it.

Before Texas Instruments bought Burr-Brown in 2000 and erased their legacy, they were a premium analog and mixed-signal manufacturer similar to Linear Technology but originating from the time of Analog Devices and Analogic.  Burr-Brown was an early player in the market for high end audio ADCs, DACs, operational amplifiers, and other functions.  They published a rich assortment of application notes many or most of which are no longer available.

[paulca]

Before Texas Instruments bought Burr-Brown in 2000 and erased their legacy, they were a premium analog and mixed-signal manufacturer similar to Linear Technology but originating from the time of Analog Devices and Analogic.  Burr-Brown was an early player in the market for high end audio ADCs, DACs, operational amplifiers, and other functions.  They published a rich assortment of application notes many or most of which are no longer available.

I think this is one of the re-incarnations today:  http://www.box-designs.com/index.php

When you poke about at several of their amps and look up the datasheets they still say Burr-Brown on the headers.

My wallet ran away and hid in the corner when I browsed that site.

[quietcat]

I doubt it. The ratios between the signal seen at the amplifier and speaker are: R_CABLE = 0.6R = 4/4.6 = 0.8696, R_CABLE = 0.3R 4/4.3 = 0.93, which may seem significant, but human hearing is logarithmic, so it makes more sense to look at it in decibels:  log₁₀(0.8696) = -0.061dB,  log₁₀(0.93) = -0.0315dB, a difference of just 0.03dB. The difference in, audio levels on each track, the sensitivity between the left and right ear, the position of the listener in the room and acoustics will easily exceed a 0.03dB difference in attenuation between channels, caused by the cable.

You forgot to multiply by 20 to get the dB, so 0.6 dB may or may not be noticeable, depending on the situation.

[6PTsocket]

I agree that any decent amp has specs that far excede the distortion of the speakers that follow it. On the subject of soft THD, you are confusing an amp used for performance with one used for playback. Some musicians intentiinally overdrive tube amps beyond their linear range because the like the effect. Solid state amps clip sharply when the limit is reached and it sounds harsh and unpleasant. Tubes clip more gradually and some musicians use it as an audio effect. In the other hand, if you are trying to reproduce music, recordrd clean or intentionally distorted, you want a clean reproduction. You do not want to distort it again and add your own special effects. Therefore, whether tube or solid state you want to stay out of the clipped signal region and use an amp with enough power to drive the speakers, which also must be sufficiently rated, to the desired level without overdriving them.

Sent from my SM-G900V using Tapatalk

[Zero999]

I doubt it. The ratios between the signal seen at the amplifier and speaker are: R_CABLE = 0.6R = 4/4.6 = 0.8696, R_CABLE = 0.3R 4/4.3 = 0.93, which may seem significant, but human hearing is logarithmic, so it makes more sense to look at it in decibels:  log₁₀(0.8696) = -0.061dB,  log₁₀(0.93) = -0.0315dB, a difference of just 0.03dB. The difference in, audio levels on each track, the sensitivity between the left and right ear, the position of the listener in the room and acoustics will easily exceed a 0.03dB difference in attenuation between channels, caused by the cable.

You forgot to multiply by 20 to get the dB, so 0.6 dB may or may not be noticeable, depending on the situation.

I stand corrected on my calculations but my point still stands. The example I used was extreme: one cable double the length of the other and both around 10% of the speaker impedance. Ideally, the cable should have a much lower resistance, so any differences in attenuation will be much less.

I agree that any decent amp has specs that far excede the distortion of the speakers that follow it. On the subject of soft THD, you are confusing an amp used for performance with one used for playback. Some musicians intentiinally overdrive tube amps beyond their linear range because the like the effect. Solid state amps clip sharply when the limit is reached and it sounds harsh and unpleasant. Tubes clip more gradually and some musicians use it as an audio effect. In the other hand, if you are trying to reproduce music, recordrd clean or intentionally distorted, you want a clean reproduction. You do not want to distort it again and add your own special effects. Therefore, whether tube or solid state you want to stay out of the clipped signal region and use an amp with enough power to drive the speakers, which also must be sufficiently rated, to the desired level without overdriving them.

Sent from my SM-G900V using Tapatalk

I believe this thread is about Hi-Fi amplifier, i.e. playback only. Distortion was mentioned, but the idea is to minimise it by not overdriving the amplifier.