• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Class A “class D killer” amplifier with THD less than -120dB

Status
Not open for further replies.

pma

Major Contributor
Joined
Feb 23, 2019
Messages
4,591
Likes
10,727
Location
Prague
Class A “class D killer” amplifier with THD less than -120dB

Intro


6 years ago I built a low-power (high heat) class A amplifier, with very low distortion. It had true class A output stage with local negative feedback to reduce distortion and a global feedback network across the whole amplifier circuit. The power was not high, like 2x25W/4ohm or 2x13W/8ohm. However, a lot of heat :). The sound was fantastic. I was not able to measure distortion of that amplifier with my system limits, so I took it on my colleague's AP2712 where we made some basic measurements, but too little to give a whole view. The amplifier needed about 170VA at idle (true class A has highest power consumption at idle) and it heated the living room identifiably. The power was not high enough for all my needs, so after about 2 years of operation I have dismounted the casework and used it elsewhere, leaving the amplifier boards assembled and working.

Time after time I resurrect these boards, make some small modifications and measure them. Now I was provoked in further measurements by those simple and never ending amplifier SINAD debates. As many of you know, I am not a big friend of 1kHz/5W SINAD as a measure or rating of amplifier quality. I am deeply convinced that there are much more important parameters and that talking about amplifier SINAD higher than 100dB is completely pointless. This is based on my more than 40 years long experience as a circuit designer and system designer and 50+ years of interest in audio circuits.

However, if you want it ….
First, why to be limited to just 1kHz numbers. If we want to consider the linearity, then not only mid-frequency linearity is important, but also low and high frequency linearity. The argument of inaudibility of high frequency distortion is pointless, as it creates intermodulation distortion products deep in the most audible part of the audio spectrum. Same stands for restricting of measuring bandwidth to 20kHz, yes we do not see the ultrasound products then, but again we know nothing about possible intermodulations in the audio band or in tweeters. The actual reason why the measuring range has been restricted in the past decade is a rapid entry of class D amplifiers to the market and corresponding business interests of the manufacturers, the parameters must look nice and there must be a continuous progress to justify new and new products.

Strictly technically speaking, the class D amplifiers still have not reached the parameters that are achievable with linear amplifiers. Now I am not talking solely about that 1kHz SINAD (or THD+N that has been used for decades to name the same parameter), but also about high frequency linearity, wide bandwidth with minimum phase shift in the audio band, fast rise time and clean output spectrum above audio band. The one and only advantage of class D amplifiers is reduced power consumption and thus smaller size, though the manufacturers and OEM assemblers then do not care much about proper thermal handling and it results in lower reliability of such products.

Back to that small class A amplifier. Class A (push-pull), in principle, keeps both + and – output halves always ON, supplied by current, so there is no switching distortion we may see in class AB, class B and class D. The biggest contributor to amplifier nonlinearity is excluded.

The PCB with the amp module looks like this

D-killer.JPG


Oscilloscope measurements

Oscilloscope measurements are a key for every circuit designer and builder even in audio. Maybe not that much for the reviewers, as we can see. Understanding of oscilloscope measurements is important, as well as lab experience. These measurements tell us about amplifier stability and help us to identify possible issues inside the amplifier circuitry, which is impossible to be done by another measuring method. The king of oscilloscope measurements are square wave measurements, which tell us very quickly if the amplifier is stable, if it has flat frequency response and if it is flat at lowest and highest frequencies. From the square response (step response) rise time and fall time we read the -3dB high frequency cut off, from the pulse tilt at lower frequency we read the low frequency cut off. From the step response shape and overshoot we read the reserve in stability or tendency to oscillations. We can see all of this in 2 minutes if we are experienced operators.

Some oscilloscope measurements on the class A module

classA_10kHz_sqresponse.png


10kHz square response is used very often in audio reviews. The plot shows input signal (blue) and amplifier output (red). We can see that there is no overshoot and the response is smooth.

We can switch the time base to see rising edge in better resolution

classA_stepresponse.png


Now we read rise time of the input square edge is 2.3us, of the output edge it is 2.64us. From these numbers we calculate amplifier rise time as 1.296us. From 1.296us rise time we calculate -3dB bandwidth as 270kHz. Easy.

Let's see the 100Hz square response

classA_100Hz_sqresponse.png


Almost no tilt of the output wave, so the low frequency extension goes well below 2Hz. We can read Gain = 28.08/1.26 = 22.29 (26.96dB)

And the 100Hz square response spectrum

classA_100Hz_spectrum.png


Another way to see where the frequency response starts to fall at high frequencies. So we have a nice 2Hz – 200kHz amplifier, a good old standard to be 10x better than is the band of interest, to minimize time and phase errors. Where have all that requests gone.

So now to the frequency domain view

Let's see the THD vs. output power at frequencies from 1kHz to 15kHz, measured with 45kHz measuring bandwidth (REW). Sorry I would not show THD+N as well, because the “+N” component is in case of this amplifier totally defined by my measuring chain noise and this would confuse the issue and would not allow to see the distortion, which would be completely buried in noise at all levels.

classA_1k-15k_thdpower.png


Still, the distortion is rather a system distortion than amplifier distortion. Even in such case we can see 1kHz THD going below -120dB and 15kHz THD between -100dB and -110dB, show mw a class D amp which can do it when measured with @45kHz BW.

Any measurement of similar kind should be verified by another method, to exclude possible systematic error. So here is the measurement made with ARTA. ARTA is unable to calibrate X-axis in [W], so you will see voltage axis [V]. Both ARTA and REW results are closely similar.

classA_thdlevel1-15k.png


Conclusion

Class D approaches to linear amplifiers in low and mid frequency linearity. We can still make linear amplifiers with much wider bandwidth, cleaner spectrum, shorter rise time and better step/square response than any class D amplifier. There is still what to investigate in class D and it should not be the 1kHz SINAD race only.
 

dadregga

Active Member
Joined
May 27, 2021
Messages
154
Likes
340
Conclusion

Class D approaches to linear amplifiers in low and mid frequency linearity. We can still make linear amplifiers with much wider bandwidth, cleaner spectrum, shorter rise time and better step/square response than any class D amplifier. There is still what to investigate in class D and it should not be the 1kHz SINAD race only.

Manufacturing cost?

Heat generation?

Efficiency/power consumption?

I'm not sure anyone is saying you can't handcraft a Class A that measures better. I think what they are saying is that Class Ds can measure as good in the audible spectrum and as good in the measurable spectrum than all but the best A/AB amps (yours measures well, nicely done), and are overall cheaper to make and vastly more efficient than either of those. Especially if you need nontrivial amounts of power.
 

Sheriff1972

Member
Forum Donor
Joined
May 29, 2019
Messages
57
Likes
41
Location
United Kingdom
Class A “class D killer” amplifier with THD less than -120dB

Intro


6 years ago I built a low-power (high heat) class A amplifier, with very low distortion. It had true class A output stage with local negative feedback to reduce distortion and a global feedback network across the whole amplifier circuit. The power was not high, like 2x25W/4ohm or 2x13W/8ohm. However, a lot of heat :). The sound was fantastic. I was not able to measure distortion of that amplifier with my system limits, so I took it on my colleague's AP2712 where we made some basic measurements, but too little to give a whole view. The amplifier needed about 170VA at idle (true class A has highest power consumption at idle) and it heated the living room identifiably. The power was not high enough for all my needs, so after about 2 years of operation I have dismounted the casework and used it elsewhere, leaving the amplifier boards assembled and working.

Time after time I resurrect these boards, make some small modifications and measure them. Now I was provoked in further measurements by those simple and never ending amplifier SINAD debates. As many of you know, I am not a big friend of 1kHz/5W SINAD as a measure or rating of amplifier quality. I am deeply convinced that there are much more important parameters and that talking about amplifier SINAD higher than 100dB is completely pointless. This is based on my more than 40 years long experience as a circuit designer and system designer and 50+ years of interest in audio circuits.

However, if you want it ….
First, why to be limited to just 1kHz numbers. If we want to consider the linearity, then not only mid-frequency linearity is important, but also low and high frequency linearity. The argument of inaudibility of high frequency distortion is pointless, as it creates intermodulation distortion products deep in the most audible part of the audio spectrum. Same stands for restricting of measuring bandwidth to 20kHz, yes we do not see the ultrasound products then, but again we know nothing about possible intermodulations in the audio band or in tweeters. The actual reason why the measuring range has been restricted in the past decade is a rapid entry of class D amplifiers to the market and corresponding business interests of the manufacturers, the parameters must look nice and there must be a continuous progress to justify new and new products.

Strictly technically speaking, the class D amplifiers still have not reached the parameters that are achievable with linear amplifiers. Now I am not talking solely about that 1kHz SINAD (or THD+N that has been used for decades to name the same parameter), but also about high frequency linearity, wide bandwidth with minimum phase shift in the audio band, fast rise time and clean output spectrum above audio band. The one and only advantage of class D amplifiers is reduced power consumption and thus smaller size, though the manufacturers and OEM assemblers then do not care much about proper thermal handling and it results in lower reliability of such products.

Back to that small class A amplifier. Class A (push-pull), in principle, keeps both + and – output halves always ON, supplied by current, so there is no switching distortion we may see in class AB, class B and class D. The biggest contributor to amplifier nonlinearity is excluded.

The PCB with the amp module looks like this

View attachment 201560

Oscilloscope measurements

Oscilloscope measurements are a key for every circuit designer and builder even in audio. Maybe not that much for the reviewers, as we can see. Understanding of oscilloscope measurements is important, as well as lab experience. These measurements tell us about amplifier stability and help us to identify possible issues inside the amplifier circuitry, which is impossible to be done by another measuring method. The king of oscilloscope measurements are square wave measurements, which tell us very quickly if the amplifier is stable, if it has flat frequency response and if it is flat at lowest and highest frequencies. From the square response (step response) rise time and fall time we read the -3dB high frequency cut off, from the pulse tilt at lower frequency we read the low frequency cut off. From the step response shape and overshoot we read the reserve in stability or tendency to oscillations. We can see all of this in 2 minutes if we are experienced operators.

Some oscilloscope measurements on the class A module

View attachment 201561

10kHz square response is used very often in audio reviews. The plot shows input signal (blue) and amplifier output (red). We can see that there is no overshoot and the response is smooth.

We can switch the time base to see rising edge in better resolution

View attachment 201562

Now we read rise time of the input square edge is 2.3us, of the output edge it is 2.64us. From these numbers we calculate amplifier rise time as 1.296us. From 1.296us rise time we calculate -3dB bandwidth as 270kHz. Easy.

Let's see the 100Hz square response

View attachment 201563

Almost no tilt of the output wave, so the low frequency extension goes well below 2Hz. We can read Gain = 28.08/1.26 = 22.29 (26.96dB)

And the 100Hz square response spectrum

View attachment 201565

Another way to see where the frequency response starts to fall at high frequencies. So we have a nice 2Hz – 200kHz amplifier, a good old standard to be 10x better than is the band of interest, to minimize time and phase errors. Where have all that requests gone.

So now to the frequency domain view

Let's see the THD vs. output power at frequencies from 1kHz to 15kHz, measured with 45kHz measuring bandwidth (REW). Sorry I would not show THD+N as well, because the “+N” component is in case of this amplifier totally defined by my measuring chain noise and this would confuse the issue and would not allow to see the distortion, which would be completely buried in noise at all levels.

View attachment 201566

Still, the distortion is rather a system distortion than amplifier distortion. Even in such case we can see 1kHz THD going below -120dB and 15kHz THD between -100dB and -110dB, show mw a class D amp which can do it when measured with @45kHz BW.

Any measurement of similar kind should be verified by another method, to exclude possible systematic error. So here is the measurement made with ARTA. ARTA is unable to calibrate X-axis in [W], so you will see voltage axis [V]. Both ARTA and REW results are closely similar.

View attachment 201567

Conclusion

Class D approaches to linear amplifiers in low and mid frequency linearity. We can still make linear amplifiers with much wider bandwidth, cleaner spectrum, shorter rise time and better step/square response than any class D amplifier. There is still what to investigate in class D and it should not be the 1kHz SINAD race only.
I would like to hear an amp like this on some of our Heritage loudspeakers. Nice work
 

MakeMineVinyl

Major Contributor
Joined
Jun 5, 2020
Messages
3,558
Likes
5,871
Location
Santa Fe, NM
Class A “class D killer” amplifier with THD less than -120dB

Intro


6 years ago I built a low-power (high heat) class A amplifier, with very low distortion. It had true class A output stage with local negative feedback to reduce distortion and a global feedback network across the whole amplifier circuit. The power was not high, like 2x25W/4ohm or 2x13W/8ohm. However, a lot of heat :). The sound was fantastic. I was not able to measure distortion of that amplifier with my system limits, so I took it on my colleague's AP2712 where we made some basic measurements, but too little to give a whole view. The amplifier needed about 170VA at idle (true class A has highest power consumption at idle) and it heated the living room identifiably. The power was not high enough for all my needs, so after about 2 years of operation I have dismounted the casework and used it elsewhere, leaving the amplifier boards assembled and working.

Time after time I resurrect these boards, make some small modifications and measure them. Now I was provoked in further measurements by those simple and never ending amplifier SINAD debates. As many of you know, I am not a big friend of 1kHz/5W SINAD as a measure or rating of amplifier quality. I am deeply convinced that there are much more important parameters and that talking about amplifier SINAD higher than 100dB is completely pointless. This is based on my more than 40 years long experience as a circuit designer and system designer and 50+ years of interest in audio circuits.

However, if you want it ….
First, why to be limited to just 1kHz numbers. If we want to consider the linearity, then not only mid-frequency linearity is important, but also low and high frequency linearity. The argument of inaudibility of high frequency distortion is pointless, as it creates intermodulation distortion products deep in the most audible part of the audio spectrum. Same stands for restricting of measuring bandwidth to 20kHz, yes we do not see the ultrasound products then, but again we know nothing about possible intermodulations in the audio band or in tweeters. The actual reason why the measuring range has been restricted in the past decade is a rapid entry of class D amplifiers to the market and corresponding business interests of the manufacturers, the parameters must look nice and there must be a continuous progress to justify new and new products.

Strictly technically speaking, the class D amplifiers still have not reached the parameters that are achievable with linear amplifiers. Now I am not talking solely about that 1kHz SINAD (or THD+N that has been used for decades to name the same parameter), but also about high frequency linearity, wide bandwidth with minimum phase shift in the audio band, fast rise time and clean output spectrum above audio band. The one and only advantage of class D amplifiers is reduced power consumption and thus smaller size, though the manufacturers and OEM assemblers then do not care much about proper thermal handling and it results in lower reliability of such products.

Back to that small class A amplifier. Class A (push-pull), in principle, keeps both + and – output halves always ON, supplied by current, so there is no switching distortion we may see in class AB, class B and class D. The biggest contributor to amplifier nonlinearity is excluded.

The PCB with the amp module looks like this

View attachment 201560

Oscilloscope measurements

Oscilloscope measurements are a key for every circuit designer and builder even in audio. Maybe not that much for the reviewers, as we can see. Understanding of oscilloscope measurements is important, as well as lab experience. These measurements tell us about amplifier stability and help us to identify possible issues inside the amplifier circuitry, which is impossible to be done by another measuring method. The king of oscilloscope measurements are square wave measurements, which tell us very quickly if the amplifier is stable, if it has flat frequency response and if it is flat at lowest and highest frequencies. From the square response (step response) rise time and fall time we read the -3dB high frequency cut off, from the pulse tilt at lower frequency we read the low frequency cut off. From the step response shape and overshoot we read the reserve in stability or tendency to oscillations. We can see all of this in 2 minutes if we are experienced operators.

Some oscilloscope measurements on the class A module

View attachment 201561

10kHz square response is used very often in audio reviews. The plot shows input signal (blue) and amplifier output (red). We can see that there is no overshoot and the response is smooth.

We can switch the time base to see rising edge in better resolution

View attachment 201562

Now we read rise time of the input square edge is 2.3us, of the output edge it is 2.64us. From these numbers we calculate amplifier rise time as 1.296us. From 1.296us rise time we calculate -3dB bandwidth as 270kHz. Easy.

Let's see the 100Hz square response

View attachment 201563

Almost no tilt of the output wave, so the low frequency extension goes well below 2Hz. We can read Gain = 28.08/1.26 = 22.29 (26.96dB)

And the 100Hz square response spectrum

View attachment 201565

Another way to see where the frequency response starts to fall at high frequencies. So we have a nice 2Hz – 200kHz amplifier, a good old standard to be 10x better than is the band of interest, to minimize time and phase errors. Where have all that requests gone.

So now to the frequency domain view

Let's see the THD vs. output power at frequencies from 1kHz to 15kHz, measured with 45kHz measuring bandwidth (REW). Sorry I would not show THD+N as well, because the “+N” component is in case of this amplifier totally defined by my measuring chain noise and this would confuse the issue and would not allow to see the distortion, which would be completely buried in noise at all levels.

View attachment 201566

Still, the distortion is rather a system distortion than amplifier distortion. Even in such case we can see 1kHz THD going below -120dB and 15kHz THD between -100dB and -110dB, show mw a class D amp which can do it when measured with @45kHz BW.

Any measurement of similar kind should be verified by another method, to exclude possible systematic error. So here is the measurement made with ARTA. ARTA is unable to calibrate X-axis in [W], so you will see voltage axis [V]. Both ARTA and REW results are closely similar.

View attachment 201567

Conclusion

Class D approaches to linear amplifiers in low and mid frequency linearity. We can still make linear amplifiers with much wider bandwidth, cleaner spectrum, shorter rise time and better step/square response than any class D amplifier. There is still what to investigate in class D and it should not be the 1kHz SINAD race only.
Class D amplification is to Class A what cassettes are to reel to reel tape at 15ips. :oops:
 
OP
pma

pma

Major Contributor
Joined
Feb 23, 2019
Messages
4,591
Likes
10,727
Location
Prague
FR in audio band
classA_FR.png



-3dB cut off frequency
classA_-3dB.png

Manufacturing cost?

Heat generation?

Efficiency/power consumption?

I'm not sure anyone is saying you can't handcraft a Class A that measures better. I think what they are saying is that Class Ds can measure as good in the audible spectrum and as good in the measurable spectrum than all but the best A/AB amps (yours measures well, nicely done), and are overall cheaper to make and vastly more efficient than either of those. Especially if you need nontrivial amounts of power.

You can make it also same good in class AB (see Benchmark AHB2). However, you have to be a real circuit designer and have production capability, in contrast of being a mere OEM modules assembler. Re price, $4k - $5k for a very good stereo class AB amplifier in small lots, half of the price for larger production quantities.
 

voodooless

Grand Contributor
Forum Donor
Joined
Jun 16, 2020
Messages
10,226
Likes
17,806
Location
Netherlands
That thing looks pretty inconspicuous :D

Here is an older advanced Class A design from 2007, achieving 110 SINAD with more power. It was posted in the Dutch version of Elektor back in the day.
 
OP
pma

pma

Major Contributor
Joined
Feb 23, 2019
Messages
4,591
Likes
10,727
Location
Prague
Distortion spectrum. Please again forget the THD+N number, +N comes from the measuring system and not from the amp. Unfortunately THD is a measuring system limit as well. Please note very low level of mains related components, though it is supplied from a conventional bridge+capacitors linear supply.

classA_1k_17W.png
 

voodooless

Grand Contributor
Forum Donor
Joined
Jun 16, 2020
Messages
10,226
Likes
17,806
Location
Netherlands
What’s the function of the opamp? That LT1122 doesn’t look like an “audio opamp” and distortion figures appear to be nowhere near your measurements.

Could the design be expanded to more power, possibly Class AB with a bit less A power? How much would performance suffer.
 
OP
pma

pma

Major Contributor
Joined
Feb 23, 2019
Messages
4,591
Likes
10,727
Location
Prague
The opamp is the opamp exactly as you would expect. LT datasheet figures are probably old measurements and underestimated. LT1122 and OPA637 give the best results in this amplifier. The opamp must be fast with high SR and low distortion. With e.g. OPA134 you get worse high frequency linearity, because of lower SR and HF loopgain of OPA134. Bipolar input opamps are unsuitable for this amp. I need low input bias current.

6 years ago the amp was measured on AP2712 and it measured H3 of -129dB. But there was less than optimal gnd track on the PCB resulting in higher mains components.

02 L-2.png
 

Hipocrates

Member
Joined
Dec 31, 2018
Messages
83
Likes
89
Location
Costa Rica
This come in actual high-powered form?

It might, but since is a class A amp, it would need a massive power supply, some parallel outputs and a huge heatsinks ''a la'' PassLabs or maybe something like the new Sugden amps with switching power supplies for class A.

To understand what's with this this amp the designer tell us

" Class A (push-pull), in principle, keeps both + and – output halves always ON, supplied by current, so there is no switching distortion we may see in class AB, class B and class D. The biggest contributor to amplifier nonlinearity is excluded."

That topology comes with some compromises.

I've experienced class A amps Sudgen, Passlabs and bedinnis paired with some ATC or Martin Logan CLS speakers (very demanding speakers) with very dynamic content, and they perform extremely well.
So regarding this amp, I can see it paired some very nice efficient speakers and for a superb stereo sound (critical listening application?), not so much the typical desktop/HT/convenience application so common on this day and age.
 

Mnyb

Major Contributor
Forum Donor
Joined
Aug 14, 2019
Messages
2,641
Likes
3,617
Location
Sweden, Västerås
I see this as a technology demonstration , even PMA say in the first post that he uses something else with slightly less absolute performance but more power in his actual music listening for obvious reasons.

With today’s inefficient loudspeakers a real class A amps is just not feasible ( there are some really hot class AB amps that claims they are class A , a contagious topic as for a while it was a marketing thing ).

On topic again , amateur question if it’s allowed .

The phase shift or rather lack of phase shift . Is this purely a function of the bandwidth of the amp or are there other factors that contributes ? Just wondering if phaseshift is a price to pay for bandwidth limiting , I suspect it is the case ?
 

Bjorn

Major Contributor
Audio Company
Forum Donor
Joined
Feb 22, 2017
Messages
1,286
Likes
2,562
Location
Norway
First, why to be limited to just 1kHz numbers. If we want to consider the linearity, then not only mid-frequency linearity is important, but also low and high frequency linearity. The argument of inaudibility of high frequency distortion is pointless, as it creates intermodulation distortion products deep in the most audible part of the audio spectrum. Same stands for restricting of measuring bandwidth to 20kHz, yes we do not see the ultrasound products then, but again we know nothing about possible intermodulations in the audio band or in tweeters. The actual reason why the measuring range has been restricted in the past decade is a rapid entry of class D amplifiers to the market and corresponding business interests of the manufacturers, the parameters must look nice and there must be a continuous progress to justify new and new products.

Strictly technically speaking, the class D amplifiers still have not reached the parameters that are achievable with linear amplifiers. Now I am not talking solely about that 1kHz SINAD (or THD+N that has been used for decades to name the same parameter), but also about high frequency linearity, wide bandwidth with minimum phase shift in the audio band, fast rise time and clean output spectrum above audio band. The one and only advantage of class D amplifiers is reduced power consumption and thus smaller size, though the manufacturers and OEM assemblers then do not care much about proper thermal handling and it results in lower reliability of such products.
Can you show measurements of intermodulation distortion and preferably also some objective listenings tests that support your claim?
 

garbz

Active Member
Joined
Sep 17, 2021
Messages
120
Likes
183
Why can't we be happy with making something of objectively good quality without the hyperbole of "Class X killer"? This hasn't killed anything. It has objectively good measurements for some situations. Class-D objectively would out perform it in others. You want to "kill" something, then scale this up to 500Wpc and produce it for under $1000.

Every amplifier class and topology has pros and cons with different purposes. We shouldn't pretend that any of them "kill" any other as precisely none of them are the best at everything which is why different ones exist in the first place.

Anyway that aside, beautiful work.

Why is KK1 and KK2 not populated? Did they end up not needing those heatsinks?
 
Status
Not open for further replies.
Top Bottom