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

[BrianHG]

The distortion does rise at higher frequencies, but I don't see what all the fuss is about? TI don't seem to have manipulated the figures at all!

I admit I missed the fact the figures I quoted were only for 20Hz, but plenty of data is provided for higher frequencies.  Refer to figure 18: +/-35V, 50W into an 8 Ohm load, THD+N <0.01% <10kHz, rising to slightly above 0.013% at 20kHz, but no one will notice the harmonic distortion above 10kHz because it will be ultrasonic. The situation (figure 16) is worse with a 4R load but still <0.03% THD+N to 20kHz. Graphs are also given at 20kHz, see figures 25 to 27, which is helpful, but beyond the audio band.

If anything TI have made the figures look worse than any one can hear. The THD+N tests were conducted with a bandwidth of 80kHz, which is excessive, because it will show a lot of ultrasonic harmonics, which no one will hear. If the bandwidth were lowered to actual audio range of 20kHz, it will more accurately reflect what can be heard, rather than what can be easily measured.

Ok, I did not mean that TI messed up the figures.  I'm saying that combine everything together, and the first paragraph printed on line page 1 description paragraph #1 says:

The LM3886 is a high-performance audio power
amplifier capable of delivering 68W of continuous
average power to a 4? load and 38W into 8? with
0.1% THD+N from 20Hz–20kHz.

This is not a lie.  It's the first thing written right there!  In the .pdf data sheet.
Your distortion charts are under controlled circumstances with a controlled single signal.

You still haven't figured out why the distortion is going down instead of up with power at the very low frequencies.  You are missing something you actually know about which is important and you just need to flex that grey matter to figure out why.  (I provided a hint in my first post pointing out what's going on...) Once you do so, you will know the consequences on the sound output driving a speaker load and I'll show you easily how this error makes an audible effect.

Actually now that I think about it, you are missing 2 things that are important...  (I'll be back when I wake up & see what you or anybody else have come up with...)

[Electro Fan]

Just want to point out, that I don't think any one here meant to say that all amplifiers sound the same, full stop. The truth is, that when an amplifier is poorly designed or operated out of specification, i.e. clipping, then distortion and noticeable differences between designs, will occur.

Roger that, and Agreed

The point is, the difference between well-designed amplifiers, operated within their specifications (no clipping), will be inaudible.

Might depend on the definition of a "well designed amplifier" but in any event you are certainly entitled to your opinion.

Want to build a reasonably priced, audibly transparent amplifier? DO NOT USE VALVES/TUBES! They're expensive and have high distortion. AVOID PASSING THE AUDIO SIGNAL THROUGH A TRANSFORMER! Audio transformers are expensive and bugger up the signal more than any decent amplifier.

Not sure what the definition of "reasonably priced" is, but DO NOT USE TUBES and AVOID OUTPUT TRANSFORMERS?  Really? 

[Zero999]

Just want to point out, that I don't think any one here meant to say that all amplifiers sound the same, full stop. The truth is, that when an amplifier is poorly designed or operated out of specification, i.e. clipping, then distortion and noticeable differences between designs, will occur.

Roger that, and Agreed

The point is, the difference between well-designed amplifiers, operated within their specifications (no clipping), will be inaudible.

Might depend on the definition of a "well designed amplifier" but in any event you are certainly entitled to your opinion.

Want to build a reasonably priced, audibly transparent amplifier? DO NOT USE VALVES/TUBES! They're expensive and have high distortion. AVOID PASSING THE AUDIO SIGNAL THROUGH A TRANSFORMER! Audio transformers are expensive and bugger up the signal more than any decent amplifier.

Not sure what the definition of "reasonably priced" is, but DO NOT USE TUBES and AVOID OUTPUT TRANSFORMERS?  Really?

I agree, reasonably priced is a subjective statement, but I stand by my comment about tubes and audio transformers being incompatible with a low distortion amplifier. Transformers are expensive, big and bulky things which introduce plenty of distortion and getting one to work down to 20Hz, requires it to be three times as big as a 60Hz mains transformer, of the same power rating as well as compromising on high frequency response. Solid state devices avoid the need for an output transformer and achieve less distortion.

The distortion does rise at higher frequencies, but I don't see what all the fuss is about? TI don't seem to have manipulated the figures at all!

I admit I missed the fact the figures I quoted were only for 20Hz, but plenty of data is provided for higher frequencies.  Refer to figure 18: +/-35V, 50W into an 8 Ohm load, THD+N <0.01% <10kHz, rising to slightly above 0.013% at 20kHz, but no one will notice the harmonic distortion above 10kHz because it will be ultrasonic. The situation (figure 16) is worse with a 4R load but still <0.03% THD+N to 20kHz. Graphs are also given at 20kHz, see figures 25 to 27, which is helpful, but beyond the audio band.

If anything TI have made the figures look worse than any one can hear. The THD+N tests were conducted with a bandwidth of 80kHz, which is excessive, because it will show a lot of ultrasonic harmonics, which no one will hear. If the bandwidth were lowered to actual audio range of 20kHz, it will more accurately reflect what can be heard, rather than what can be easily measured.

Ok, I did not mean that TI messed up the figures.  I'm saying that combine everything together, and the first paragraph printed on line page 1 description paragraph #1 says:

The LM3886 is a high-performance audio power
amplifier capable of delivering 68W of continuous
average power to a 4? load and 38W into 8? with
0.1% THD+N from 20Hz–20kHz.

This is not a lie.  It's the first thing written right there!  In the .pdf data sheet.
Your distortion charts are under controlled circumstances with a controlled single signal.

You still haven't figured out why the distortion is going down instead of up with power at the very low frequencies.  You are missing something you actually know about which is important and you just need to flex that grey matter to figure out why.  (I provided a hint in my first post pointing out what's going on...) Once you do so, you will know the consequences on the sound output driving a speaker load and I'll show you easily how this error makes an audible effect.

Actually now that I think about it, you are missing 2 things that are important...  (I'll be back when I wake up & see what you or anybody else have come up with...)

At lower power levels there will be more crossover distortion and noise, relative to the signal. Is that what you're referring to? Yes I did miss that, but still isn't bad. Extrapolating the the worst graph, figure 25, shows only 0.06% THD+N at 10mW (hardly audible at any frequency through an inefficient, high quality speaker, let alone, at the ultrasonic frequency of 20kHz).

That needs to be put into perspective. What sort of dynamic range does the source have? Suppose it's a CD (vinyl will be much worse) and you've got the volume set to a reasonable level, yet the output is low because a quiet part of the music is playing, there'll be more distortion due to the quantisation error of the DAC, than the amplifier, at those low power levels.

[Fungus]

What are the best amp modules for use with external line level audio, like from a computer?

I would like to upgrade my computer's sound with a clean powerful quiet audio module that can use my power supply and drive good external speakers.. like the small Dayton Audio ones without distortion.

Me? I'd get a USB sound output gadget and some active speakers.

USB sound is easy: https://www.ebay.com/itm/291614747261

(Randomly chosen seller, search for [img=https://www.ebay.com/sch/i.html?_nkw=pcm2704]http://PCM2704[/img] to find more like it)

I own three PCM2704 devices and the sound from them is impeccable.

For under $10 you should have plenty left over for some nice speakers.

[Buriedcode]

Just wanted to second the USB soundcard option.  Whilst it'll most likely have more jitter than an on-board/built-in soundcard (since the sample rate comes off a PLL driven by USB timing) it isn't noticeable, but the fact its 'not in the PC' reduces noise.  This doesn't' seem to be much of a problem these days, but there is a definite difference between internal and external when recording.

Even the $1 CM108 USB soundcard dongles sound pretty good.

[BrianHG]

At lower power levels there will be more crossover distortion and noise, relative to the signal. Is that what you're referring to? Yes I did miss that, but still isn't bad. Extrapolating the the worst graph, figure 25, shows only 0.06% THD+N at 10mW (hardly audible at any frequency through an inefficient, high quality speaker, let alone, at the ultrasonic frequency of 20kHz).

That needs to be put into perspective. What sort of dynamic range does the source have? Suppose it's a CD (vinyl will be much worse) and you've got the volume set to a reasonable level, yet the output is low because a quiet part of the music is playing, there'll be more distortion due to the quantisation error of the DAC, than the amplifier, at those low power levels.

  Crossover distortion, you figured it out #1.  Now, you seem to be missing it's importance and consequences with regard to music, how we hear, and signal power at different frequencies which pertains to #2.  So, I went through the trouble of setting up this little illustration & audio clip.  I deliberately chosen cheap pop-rock recording, normal CD quality, originally taken from a cheap .mp3.  No fancy soft and loud classical music with wide dynamic range.


See illustration and optional listen to 2 second clip.


As you can see, at the higher frequencies, the audio power is a fraction that of the low frequency bass drum, yet, it still sounds as loud to our ears.  In fact, looking at that full power 50 watt signal where the high hat cymbal drums mixed in, which you can hear in the audio, is recorded at full volume.  No tricks, yet, the audio power of the above 8Khz signal, well within .mp3 and 16 bit 44.1k audio, has only peaked at 0.7watts and still has definition at 0.06watts.  According to TI's specs, lower power and higher frequencies is this IC's weak spot.  (Hint: When testing aligning audio systems, Pink noise is used instead of white noise...  Now what's special about the 'Pink Noise')

What would happen if I want to lower my volume at 12 watts today, not full blast 50.  The first cymbal strike would go from 0.7w down to 0.17w and the second would be down at 0.015w.  (I can still clearly hear both High Hat cymbals in this song on my equipment at 12w, 50w and full 250w, though I don't own any audio equipment which shows any cross-over distortion on my scope...)

Remember, crossover distortion is not a slight smooth angular bend or deformation along the full signal voltage in the output signal vs the input signal.  It is 2 small 'voids' where neither the NPN or PNP output transistors are driving the output with anything for a short time making a flat linear void where these small wattage signals can sit inside.  For the most part, depending on how bad the crossover distortion is and how temperature of the IC amp usually shifts that about makes this IC great for box radios, some TVs, ghetto-blasters, compact all in 1 systems, but I find this un-acceptable for high fidelity solutions.  Especially that enough Class B and Class AB designs exist which can do better, except that the ones without such distortion tend to get hot and need a big heatsink.  And if you want peak minimal distortion at low power, and a bit of linearity loss at full power, but still no sign of any small signal being chewed up by the big bass drum strikes, Class A amps are the way to go.

This lastly brings me to Class D amps.  In theory, a properly designed Class D amp shouldn't have this problem.  In an attempt to visualize the working mechanics of a class D amp, The crossover problems should be gone, but, what seems to happen is as the output power goes up, refinement transfer of the source signal does get sloppy, or worse.  I do have a few Ideas on how to design a clean musical class D amp at all power levels, but, it means using expensive GAN fets for speed and low rds figures and FPGA components and building everything from scratch the amp would only have digital in.

[BrianHG]

I wonder If I tried my classical music CDs as an example.  One good symphony bass drum strike going down to a small quiet section with flutes and chimes.  The power difference would be 50 watts to something like 0.1w in the quiet section's higher frequency tones...  I choose not to because I realized that would not be fair for mainstream normal pop music recorded mainly all at one volume and it wouldn't even show on the screen capture.  You would only get a single pixel wiggling up and down during the quite passage.

[alexanderbrevig]

What makes an amp better?

This is my list:
Appropriate wattage and headroom for its application.
Sane thermal shutdown.
Can handle abuse even if handled by ignorant voluntary 'stage hands' or visitors.
THD an order of magnitude less than the speakers it's driving.
Appropriate frequency response for it's intended load.
Bridgeable for double power, easily configurable.
Phase linear DSP EQ.
A good power supply (yes, most of them has more than sufficient PSU but I've encountered cheapo Behringers and Peavy, they got noisy when sharing grid with incandescent dimmers), for home use it should reject noise imposed by things like ethernet over power grid (again, not all do for some reason).

Do amps sound different? Absolutely.
Do I hear a difference between a properly powerful Crown vs a properly powerful Crest or QSC (assuming no DSP and driving the same speakers)? Absolutely not.

[paulca]

I deliberately chosen cheap pop-rock recording, normal CD quality, originally taken from a cheap .mp3.  No fancy soft and loud classical music with wide dynamic range.

Not sure if it effects your results, but mp3 is, of course, lossful.  One of the first things it does it divide the frequency spectrum into 64 (I think) bands and throws the odd ones away, dropping the bandwidth requirements by 2.  This is called physco-acoustics where the ear will generally ignore frequencies either side of each other.  It then quantises the remaining waveform using DCT.  This is further quantized in the time domain.  The result, at high compression levels, results in quantization noise artifacts.  Thus, not only does it remove a lot of the signal at encoding time, but adds other stuff back in at decode.

An interesting experiment is to run an MP3 encode, decode, encode in a loop, so you encode, decode and re-encode a sample multiple times.  Surprisingly it will eventually reach a stable state where it cannot be quantized further and subsequent re-encodes lose nothing more.

[BrianHG]

I deliberately chosen cheap pop-rock recording, normal CD quality, originally taken from a cheap .mp3.  No fancy soft and loud classical music with wide dynamic range.

Not sure if it effects your results, but mp3 is, of course, lossful.  One of the first things it does it divide the frequency spectrum into 64 (I think) bands and throws the odd ones away, dropping the bandwidth requirements by 2.  This is called physco-acoustics where the ear will generally ignore frequencies either side of each other.  It then quantises the remaining waveform using DCT.  This is further quantized in the time domain.  The result, at high compression levels, results in quantization noise artifacts.  Thus, not only does it remove a lot of the signal at encoding time, but adds other stuff back in at decode.

An interesting experiment is to run an MP3 encode, decode, encode in a loop, so you encode, decode and re-encode a sample multiple times.  Surprisingly it will eventually reach a stable state where it cannot be quantized further and subsequent re-encodes lose nothing more.

Just wanted to show that even a .mp3 can have a wide dynamic response, with soft cymbals miked in with a powerful bass drum kick, well, in the 8Khz region anyways and still can be used with a good amp.  I know .mp3 would drop out above 12khz for low bitrates, 15 khz for high bitrates when there is a good bass drum and snare drum strike where a .flac or cd original doesn't exhibit this flaw in the transcoding.

[Fungus]

Just wanted to second the USB soundcard option.  Whilst it'll most likely have more jitter than an on-board/built-in soundcard (since the sample rate comes off a PLL driven by USB timing) it isn't noticeable, but the fact its 'not in the PC' reduces noise.  This doesn't' seem to be much of a problem these days, but there is a definite difference between internal and external when recording.

A PLL might not be exactly the desired frequency but they're reasonably stable, they don't cause jitter per se.

Those TI PCM2704 chips have a "unique system that recovers the audio clock from USB packet data. On-chip analog PLLs with SpAct(tm) enable playback with low clock jitter".

Presumably that means they look at the rate the data is arriving over USB and re-tune the PLL accordingly, over time.

Whatever they're doing, it seems to work. I own three different variants of those things now, that blue one is a good choice for connecting to active speakers (and also has optical/coax, etc. if you need them).

[Zero999]

At lower power levels there will be more crossover distortion and noise, relative to the signal. Is that what you're referring to? Yes I did miss that, but still isn't bad. Extrapolating the the worst graph, figure 25, shows only 0.06% THD+N at 10mW (hardly audible at any frequency through an inefficient, high quality speaker, let alone, at the ultrasonic frequency of 20kHz).

That needs to be put into perspective. What sort of dynamic range does the source have? Suppose it's a CD (vinyl will be much worse) and you've got the volume set to a reasonable level, yet the output is low because a quiet part of the music is playing, there'll be more distortion due to the quantisation error of the DAC, than the amplifier, at those low power levels.

 :-+ Crossover distortion, you figured it out #1.  Now, you seem to be missing it's importance and consequences with regard to music, how we hear, and signal power at different frequencies which pertains to #2.  So, I went through the trouble of setting up this little illustration & audio clip.  I deliberately chosen cheap pop-rock recording, normal CD quality, originally taken from a cheap .mp3.  No fancy soft and loud classical music with wide dynamic range.


See illustration and optional listen to 2 second clip.


As you can see, at the higher frequencies, the audio power is a fraction that of the low frequency bass drum, yet, it still sounds as loud to our ears.  In fact, looking at that full power 50 watt signal where the high hat cymbal drums mixed in, which you can hear in the audio, is recorded at full volume.  No tricks, yet, the audio power of the above 8Khz signal, well within .mp3 and 16 bit 44.1k audio, has only peaked at 0.7watts and still has definition at 0.06watts.  According to TI's specs, lower power and higher frequencies is this IC's weak spot.  (Hint: When testing aligning audio systems, Pink noise is used instead of white noise...  Now what's special about the 'Pink Noise')

What would happen if I want to lower my volume at 12 watts today, not full blast 50.  The first cymbal strike would go from 0.7w down to 0.17w and the second would be down at 0.015w.  (I can still clearly hear both High Hat cymbals in this song on my equipment at 12w, 50w and full 250w, though I don't own any audio equipment which shows any cross-over distortion on my scope...)

Remember, crossover distortion is not a slight smooth angular bend or deformation along the full signal voltage in the output signal vs the input signal.  It is 2 small 'voids' where neither the NPN or PNP output transistors are driving the output with anything for a short time making a flat linear void where these small wattage signals can sit inside.  For the most part, depending on how bad the crossover distortion is and how temperature of the IC amp usually shifts that about makes this IC great for box radios, some TVs, ghetto-blasters, compact all in 1 systems, but I find this un-acceptable for high fidelity solutions.  Especially that enough Class B and Class AB designs exist which can do better, except that the ones without such distortion tend to get hot and need a big heatsink.  And if you want peak minimal distortion at low power, and a bit of linearity loss at full power, but still no sign of any small signal being chewed up by the big bass drum strikes, Class A amps are the way to go.

This lastly brings me to Class D amps.  In theory, a properly designed Class D amp shouldn't have this problem.  In an attempt to visualize the working mechanics of a class D amp, The crossover problems should be gone, but, what seems to happen is as the output power goes up, refinement transfer of the source signal does get sloppy, or worse.  I do have a few Ideas on how to design a clean musical class D amp at all power levels, but, it means using expensive GAN fets for speed and low rds figures and FPGA components and building everything from scratch the amp would only have digital in.

Also there's no mention in the data sheet of how much of it is crossover distortion, how much is due to general non-linearity and lastly noise, which will also become greater, at lower power levels, but might be more subtle.

How noticeable will it be though? And how much will the amplifier contribute to the distortion, compared to what's already there in the audio source?

Let's look at the figures, assuming 0.1% THD+N, which is worse than any of the figures given on the LM3886 graphs:

10mW_P out, 0.1% THD:
V_P@10mW_P in to 8R:
V_P = (P*R)^0.5 = (0.01*8)^0.5 = 0.2828V_P

V_P(ERROR), 0.2828V_P, 0.1% THD:
V_P(ERROR) = V_P*0.1% = 0.2828*0.001 = 282.8*10^-6V

In the example with 50W_P
V_P = (P*R)^0.5 = (50*8)^0.5 = 20V_P

Let's look at the quantisation voltage of the DAC, using a 16-bit per sample, which in this case will be used to represent the full +/-20V swing, of the amplifier.

V_PP = 2*V_P = 2*20 = 40V_PP
V_ERROR = V_PP/(2^BITS) = 40/2¹⁶ = 40/65536 = 610.35*10^-6V

This is over double the 282.8µV error of the amplifier and how many 16-bit ADCs and DACs have an error and linearity as good as 1-bit? Other factors are: the recording is unlikely to use the entire 16-bit dynamic range and the distortion in the recording equipment. The real error in the source will be worse than 610.35µV error on the output of the amplifier!

One could argue that listening at low levels, will cause the amplifier's error to dominate, but the real THD of the LM3886 is much lower than 0.1%, which was an extreme example and the quieter the sound is, the chances are greater that the errors will fall below the threshold of hearing.

I agree that class D amplifiers are not free from distortion but if it's better than what can be recorded and heard, then they're transparent and contribute no additional distortion to the sound.

Conclusion: the sound is only as good as its source and one's hearing, no matter how good the final power amplifier is and aiming to be significantly better than that, results in diminishing returns, until it's pointless.

[Alex Nikitin]

Conclusion: the sound is only as good as its source and one's hearing, no matter how good the final power amplifier is and aiming to be significantly better than that, results in diminishing returns, until it's pointless.

That is presumably correct if you are listening only to continuous sine waves, used in all quoted measurements. If that is what you do, I have no objections  .

Cheers

Alex

[macboy]

...

First of all, there are no legal standards against which audio amps are measured.

In the US there has been since 1974:

https://www.ftc.gov/enforcement/rules/rulemaking-regulatory-reform-proceedings/amplifier-rule

Certainly the FTC has set forth rules. And it even reviews and updates those rules over time. But ever since 1974 and still today, those rules are so loose and broad that there is no chance whatsoever of direct comparison between amplifiers based on the numbers produced. The FTC rules allow the THD level and the bandwidth to be at the discretion of the tester (manufacturer). This means that the el-cheapo will be tested at 1% THD and 1kHz and might have a chance of producing decently high power numbers within those limits. The respectable manufacturers will set their limits to say 0.05% and 20 Hz - 20 kHz, and a different (always lower) power number will fall out of the testing due to the more restrictive specs. Both tests meet the rules, because the arbitrarily chosen THD level and bandwidth are given alongside the power rating.

The FTC "Rules" are not "standards".

[Zero999]

The more I look into this, the more pointless aiming for ridiculously low levels of distortion and noise becomes.

How noisy is a 500R dynamic microphone?

V_N = (4kTBR)^0.5
k = 1.3803*10^-23
T = 20^oC + 273.16 = 293.16^oK
B = 20kHz
R = 500Ohm

V_N = (4kTBR)^0.5 = (4*1.3803*10^-23*293.16*20*10³*500)^0.5 = 402.32*10^-9V

Taken from the data sheet of a random dynamic microphone, the sensitivity is -72dB/Pa (94dB SPL).
V = 10^(-72/20) = 251*10^-6V
https://www.bogen.com/products/pdfs/microphonespdfs/HDO100m.pdf

Suppose we want 10V out of the amplifier, full range, when the peak sound into the microphone is 94dB.
A_V = 10/251*10^-6 = 39.81*10³

The noise, due to the 500R microphone on the amplifier's output now is:
V_N = 39.81*10³*402.32*10^-9 = 15.61mV

Much higher than the figures discussed above.

[David Hess]

Remember, crossover distortion is not a slight smooth angular bend or deformation along the full signal voltage in the output signal vs the input signal.  It is 2 small 'voids' where neither the NPN or PNP output transistors are driving the output with anything for a short time making a flat linear void where these small wattage signals can sit inside.

...

Especially that enough Class B and Class AB designs exist which can do better, except that the ones without such distortion tend to get hot and need a big heatsink.  And if you want peak minimal distortion at low power, and a bit of linearity loss at full power, but still no sign of any small signal being chewed up by the big bass drum strikes, Class A amps are the way to go.

The idle current for a class-AB design is insignificant compared to the output power.  The large heat sink is required regardless.

Good designs are closer to class-B than class-AB because of a problem different than a "void" in gain at the crossover point.  When both output transistors are conducting, gain is contributed by both so the gain (transconductance) actually increases producing a hump at zero output current.  It is not the lack of gain at crossover but the variation in gain at crossover and on either side which produce something like a flattened "W" in the transconductance graph.

A further problem is that the bias conditions change with temperature so after a period of high output power, the transistors are hotter and the bias voltage should track to maintain the same idle current.  Bias circuits are designed to track temperature but it is the temperature of the transistor dies which matters and there is a considerable time lag between them and the heat sink or even transistor cases.  Better designs mount the temperature sensor as close as possible to the transistor dies to minimize lag in the temperature measurement.  I usually epoxy the temperature sensing diode directly to the transistor's metal tabs.

In a good design, the output stage contributes a majority of the distortion because of the above difficulties and there are all kinds of clever ways to handle them.  On semiconductor and others produce transistors which have a temperature measurement diode mounted adjacent to the transistor die to minimize thermal lag.  MOSFETs have less temperature dependence to start with although this comes at the expense of lower transconductance and higher cost.  Local feedback around the output stage can reduce the magnitude of gain variation at the cost of complexity.

Simple class-AB designs can achieve 0.1% distortion worst case.  Complex ones can achieve 0.001% or better.  Considering how much distortion large signal audio transducers (speakers) produce, it is very easy to reach the point of diminishing returns as far as distortion and it might be better to spend more on better source material, a better environment, better speakers, and handling clipping or at least not making it worse.  That last thing is difficult to convey in technical specifications however and marketing is too untrustworthy to believe.

[BrianHG]

Also there's no mention in the data sheet of how much of it is crossover distortion, how much is due to general non-linearity and lastly noise, which will also become greater, at lower power levels, but might be more subtle.

How noticeable will it be though? And how much will the amplifier contribute to the distortion, compared to what's already there in the audio source?

Let's look at the figures, assuming 0.1% THD+N, which is worse than any of the figures given on the LM3886 graphs:

10mW_P out, 0.1% THD:
V_P@10mW_P in to 8R:
V_P = (P*R)^0.5 = (0.01*^0.5 = 0.2828V_P

V_P(ERROR), 0.2828V_P, 0.1% THD:
V_P(ERROR) = V_P*0.1% = 0.2828*0.001 = 282.8*10^-6V

In the example with 50W_P
V_P = (P*R)^0.5 = (50*^0.5 = 20V_P

Let's look at the quantisation voltage of the DAC, using a 16-bit per sample, which in this case will be used to represent the full +/-20V swing, of the amplifier.

V_PP = 2*V_P = 2*20 = 40V_PP
V_ERROR = V_PP/(2^BITS) = 40/2¹⁶ = 40/65536 = 610.35*10^-6V

This is over double the 282.8µV error of the amplifier and how many 16-bit ADCs and DACs have an error and linearity as good as 1-bit? Other factors are: the recording is unlikely to use the entire 16-bit dynamic range and the distortion in the recording equipment. The real error in the source will be worse than 610.35µV error on the output of the amplifier!

One could argue that listening at low levels, will cause the amplifier's error to dominate, but the real THD of the LM3886 is much lower than 0.1%, which was an extreme example and the quieter the sound is, the chances are greater that the errors will fall below the threshold of hearing.

I agree that class D amplifiers are not free from distortion but if it's better than what can be recorded and heard, then they're transparent and contribute no additional distortion to the sound.

Conclusion: the sound is only as good as its source and one's hearing, no matter how good the final power amplifier is and aiming to be significantly better than that, results in diminishing returns, until it's pointless.

Ok, we are getting closer to the same page, though we don't need such a great overall THD or linearity, since this is not how our ears work, lets take a look at a real LM3886's measured crossover distortion and it's actual consequences.  You are going overboard with respect to super low THD and super linearity into the realm of I dare say audiophoolery.  What I am showing here is that even with a great 0.01% overall THD, even in the bounds of my above audio example at both my volume choices, an audio effect is being produced which has an effect on the sound because of where it sits and how it is shaped.

Here is a 1Khz sound driven though a LM3886 driving a 6 ohm speaker:
(Scope shot taken from http://www.diyaudio.com/forums/chip-amps/312970-lm3886-crossover-hf-noise-output.html )

This distortion bend at +0.5v and -0.5v has a visible angle edge to it.  Even at full volume, during the the 'Releasing open the high hats cymbals part open', the middle 25% of that amplitude in my audio sample would play at a lower volume getting a boost once it passed that 0.5v threshold.  In fact it actually worse.  Inside, that 'cross-over' zone, since there the driving transistors aren't fully conduction, the response at higher frequencies like 8KHz may only spike through during the loudest peaks while the rest being muted even more than the example 1KHz tone.  If I lower the volume of my example volume to 12watts, 50% of the amplitude of the 'Releasing open the high hats cymbals part open' section of the music would sit in that distortion zone creating a sort of gated noise filter audio effect.  (IE, cut down the volume, especially treble, when there is too little sound to remove hiss, kick up the volume once you pass a threshold.  Though the effect may make your source audio appear to have less background noise in it than it actually has...)

Now as to your comment about achieving that unnecessary impossible THD linearity with your source dac and an amp too perfect, what if I were to tell you that I can make an amp with only 2 active components, whose only 'sin' is that the overall THD is 0.3%, much worse than the LM3886 because the output looks like this:

In this hand drawing approximation the blue represents the 0% error THD desired output, a full 50 watt signal.  The red represents the output of my lousy 0.3% THD amp.  With this cheap 2 active component amp, the distortion is curved and smoothed across the full 50 watt output.  But, now, I can guarantee that no matter circuit temperature, it is impossible to mess up the medium-small signals like in the 1KHz LM3886 scope-shot.  In fact, the middle 12 watts, even the middle 25 watts would have a THD comparable to the LM3886 and the error at the high powered edges wouldn't even be noticed by the ears since our ears have difficulty hearing fine true linear definition at high power levels, especially if the transformation is a smooth gradual transition over a wide span.  The key feature is that middle unpredictable gap in the LM3886 is impossible in my design.

[BrianHG]

@Hero999, as for your DAC THD accuracy.  If my audio sample is at full range, 16bits, =+/-20v, the high hats I measured at +/-2v will have a resolution of 13 bits.  Are you saying that a 16bit dac can't produce any reasonably definable audible sound with 13 bits.  Even the tiny 0.7v high hat cymbals release has over 11 bits of definition in it.  In fact, using my cheap Amiga 8 bit audio sound, the high hats would still contain 3 bits of definition, with a little background hiss from my cheap 8 bit sampler, it would still generate an recognized audible cymbal sound.  A 12 bit dac on a 1$ MCU today would roast that definition.  Your don't need to micromanage every single last 1 bit from an audio DAC like it is instrumentation.  The original uncompressed 16 bit CD audio if 'Phenomenal' and way beyond most of our hearing capabilities, though, it is nice to find the penultimate reference if you really want that perfect linearity over the full voltage and frequency range.  A slight linearity error from -1v through +1v leading to a lowered THD would never be noticed by the ear.

However, that 'cross-over distortion hole' in the sound, our ear can fill in some of the missing sound and you can get used to it, but, get rid of it completely, now your audio amp will have a transparent open window/air quality of sound, with real refined image at any volume with any type of reasonable audio source starting from even around 1/4 the quality of a CD and up...

[Gyro]

A further problem is that the bias conditions change with temperature so after a period of high output power, the transistors are hotter and the bias voltage should track to maintain the same idle current.  Bias circuits are designed to track temperature but it is the temperature of the transistor dies which matters and there is a considerable time lag between them and the heat sink or even transistor cases.  Better designs mount the temperature sensor as close as possible to the transistor dies to minimize lag in the temperature measurement.  I usually epoxy the temperature sensing diode directly to the transistor's metal tabs.

This is one area where integrated devices like the LM3886 actually benefit, as they have direct temperature monitoring of the output device junctions. I remember Sanken (I think) producing audio output transistors with on-die monitoring too, unfortunately heavily faked.

Already mentioned in another thread, but there's a reasonably detailed spice analysis of the LM3886 and the effect of circuit parasitics here...

https://www.neurochrome.com/taming-the-lm3886-chip-amplifier/stability/

Interesting to see the effect of fairly small layout inductances on phase margin.

[Zero999]

@Hero999, as for your DAC THD accuracy.  If my audio sample is at full range, 16bits, =+/-20v, the high hats I measured at +/-2v will have a resolution of 13 bits.  Are you saying that a 16bit dac can't produce any reasonably definable audible sound with 13 bits.  Even the tiny 0.7v high hat cymbals release has over 11 bits of definition in it.  In fact, using my cheap Amiga 8 bit audio sound, the high hats would still contain 3 bits of definition, with a little background hiss from my cheap 8 bit sampler, it would still generate an recognized audible cymbal sound.  A 12 bit dac on a 1$ MCU today would roast that definition.  Your don't need to micromanage every single last 1 bit from an audio DAC like it is instrumentation.  The original uncompressed 16 bit CD audio if 'Phenomenal' and way beyond most of our hearing capabilities, though, it is nice to find the penultimate reference if you really want that perfect linearity over the full voltage and frequency range.  A slight linearity error from -1v through +1v leading to a lowered THD would never be noticed by the ear.

I agree with you about the number of bits required for audio transparency. In many cases 8-bit per sample is enough, especially for lots of pop music which is heavily compressed to start with.

However, that 'cross-over distortion hole' in the sound, our ear can fill in some of the missing sound and you can get used to it, but, get rid of it completely, now your audio amp will have a transparent open window/air quality of sound, with real refined image at any volume with any type of reasonable audio source starting from even around 1/4 the quality of a CD and up...

I see your point about cross-over distortion.

Regarding the output waveform from the LM3886 posted in your previous post: they look pretty bad, but we can't be sure genuine parts were used. The person conducting the test ordered some LM3886s off a random ebay seller, so could be fake. I'll reserve judgement until I see an example using genuine parts.

[Alex Nikitin]

Regarding the output waveform from the LM3886 posted in your previous post: they look pretty bad, but we can't be sure genuine parts were used. The person conducting the test ordered some LM3886s off a random ebay seller, so could be fake. I'll reserve judgement until I see an example using genuine parts.

If you follow the link you'll find that these distortion were caused by an insufficient current drive on the MUTE pin. A nice picture to illustrate a concept of crossover distortion but it does not show real distortion of a properly operated LM3886.

Cheers

Alex

P.S. - just FWIW, I do not consider "CD quality" (uncompressed 16bit 44.1kHz) as sufficient for Hi-Fi audio. As a fact I am rarely (maybe once a year) listen to  this kind of quality. My CD player is off almost all the time, I listen to vinyl and tape, occasionally to hi-res digital files (24bit 96kHz).

[BrianHG]

Regarding the output waveform from the LM3886 posted in your previous post: they look pretty bad, but we can't be sure genuine parts were used. The person conducting the test ordered some LM3886s off a random ebay seller, so could be fake. I'll reserve judgement until I see an example using genuine parts.

If you follow the link you'll find that these distortion were caused by an insufficient current drive on the MUTE pin. A nice picture to illustrate a concept of crossover distortion but it does not show real distortion of a properly operated LM3886.

Cheers

Alex

Good catch.  The super tiny print of the middle paragraph explaining the solution led to me ignoring that point.  I guess the LM3886 is better that what was shown.  Now I need to find a proper setup driven to it's 50watts peak capability to see how it performs under load, not just at 1w like shown in the scopeshot.

P.S. - just FWIW, I do not consider "CD quality" (uncompressed 16bit 44.1kHz) as sufficient for Hi-Fi audio. As a fact I am rarely (maybe once a year) listen to  this kind of quality. My CD player is off almost all the time, I listen to vinyl and tape, occasionally to hi-res digital files (24bit 96kHz).

The high end CDs I have, like from Telarc Records, do appear to be well mastered.  Though I would not call them the penultimate HiFi, I cant argue that 24/96 hires audio has obviously 256x the dynamic amplitude while for most of us with 20Khz to hearing bandwidth, the extra 2x oversampling, or better 4x of 24/192 audio does produce much better defined captures of things like that initial striking hit of a drum stick on the Ride and Clash cymbals, and you can much more clearly make out the positioning and imaging of the set of springs rattling under the snare drum, which only just barely make it through on better quality CD recordings.  On regular cheap modern pop recordings, the springs are more akin to a simple white or pink noise.

[paulca]

P.S. - just FWIW, I do not consider "CD quality" (uncompressed 16bit 44.1kHz) as sufficient for Hi-Fi audio. As a fact I am rarely (maybe once a year) listen to  this kind of quality. My CD player is off almost all the time, I listen to vinyl and tape, occasionally to hi-res digital files (24bit 96kHz).

I think it depends on what you want to do with the recording.  If you want to listen to it, then 44.1k 16bit is absolutely fine.  In fact in many, many tests it has been proved that HiFi aficionados do NOT correctly identify higher rate recordings with any statistically significant margin. 

To argue against it argues against physical limitations of our ears and physco-acoustics.  Remember what you think you hear bares a lot less resemblance to what the sound actually is than you think.  It also argues against sampling theorem.  Our ears are evolved for specific purposes, hunting, defense and language, not listening to music, so they suck at the later.

The dynamic range on tape... depends on the tape, domestic tape is useless compared to high speed, wide, studio mastering tape for example.  The quality of vinyl is orders of magnitude lower than CD in particular due to noise and groove width.  To get the same range as CD you would have to make the groves very wide and well spaced and thus not get much on the record.  What vinyl users often refer to as "better sound quality and warmth" is actually just noise and imperfections not originally recorded on the medium, but added by the archaic technology.  Digital CDs are much cleaner, so if the recording is sterile and cold so will the play back be.  This means you hear it as the artist recorded it. 

Vinyl is akin to TV pictures which are deliberately blurred and persisted compared to a reference monitor, the later makes normal TV and movies look cold, sterile and ugly.  Try watching a DVD on a good high end PC monitor and you'll see what I mean.  Then watch it on a good TV.  Reverse the experiment and view your PC on your TV with the same settings.  It will look blurred and softened etc.

[Zero999]

Regarding the output waveform from the LM3886 posted in your previous post: they look pretty bad, but we can't be sure genuine parts were used. The person conducting the test ordered some LM3886s off a random ebay seller, so could be fake. I'll reserve judgement until I see an example using genuine parts.

If you follow the link you'll find that these distortion were caused by an insufficient current drive on the MUTE pin. A nice picture to illustrate a concept of crossover distortion but it does not show real distortion of a properly operated LM3886.

Cheers

Alex

Ah that makes sense. Thanks for reading the whole thread.

P.S. - just FWIW, I do not consider "CD quality" (uncompressed 16bit 44.1kHz) as sufficient for Hi-Fi audio. As a fact I am rarely (maybe once a year) listen to  this kind of quality. My CD player is off almost all the time, I listen to vinyl and tape, occasionally to hi-res digital files (24bit 96kHz).

What's wrong with CD quality? It's much lower distortion and noise than vinyl and tape: hiss, hiss, flutter, flutter, wow, wow. If you like that sort of thing then good for you, but steer clear of any digital format, unless it's ripped from vinyl or tape.

Another interesting fact is, plenty of vinyl and tape recordings will have already been through a 16-bit ADC and DAC, before you get them. Many early digital recordings were actually initially sampled at just above CD quality, 16-bit 50kHz, then transferred to tape and vinyl for release. Tape was considered to be too noisy and distorted for the master/initial recording.
https://en.wikipedia.org/wiki/Hard_disk_recorder#History

24-bit 96kHz? That's just a waste of bandwidth! Good for recording and signal processing, but not worth it for playback.

[BrianHG]

The dynamic range on tape... depends on the tape, domestic tape is useless compared to high speed, wide, studio mastering tape for example.  The quality of vinyl is orders of magnitude lower than CD in particular due to noise and groove width.  To get the same range as CD you would have to make the groves very wide and well spaced and thus not get much on the record.  What vinyl users often refer to as "better sound quality and warmth" is actually just noise and imperfections not originally recorded on the medium, but added by the archaic technology.  Digital CDs are much cleaner, so if the recording is sterile and cold so will the play back be.  This means you hear it as the artist recorded it.

Vinyl is akin to TV pictures which are deliberately blurred and persisted compared to a reference monitor, the later makes normal TV and movies look cold, sterile and ugly.  Try watching a DVD on a good high end PC monitor and you'll see what I mean.  Then watch it on a good TV.  Reverse the experiment and view your PC on your TV with the same settings.  It will look blurred and softened etc.

Error:  Blurred, Warmth??? What are you talking about?  Vinyl has a 70KHz bandwidth.  Blurred? My 180 gram vinyl copy of Supertramp's Crime of the Century has cymbals strikes (I HD sampled and verified that signals past 50KHz exist) are so sharp and crystal clear compared to the cheap 22 Khz CD version, at full volume on my audio system, which is flat to 100KHz, the playback rips my ear drums to shreds with the most strikingly razor sound I've experienced except for 192KHz sampled audio...  Don't confuse a mastering choice when the vinyl version, or those writeups in audiophoolery magazines specifically cherry picking smoothly mastered/muffled junk as examples, with a properly full bandwidth pressed vinyl is created, the sound except for the needle traveling groove noise, is astounding for a physical medium.  That being said, 24/192k digital recordings free of all noise still cremates it.