For the most part (right or wrong), I use the term peak or RMS 'power' depending on the type of voltage used in the power formula, into a purely resistive load. When I bench-test an amp for power, I measure the RMS voltage into a dummy load resistor, right at the point of clipping and call that RMS power. You are, of course, correct. There is no such thing in physics as RMS power.
For many amps, the manufacturer uses the full peak voltage, no load to 'calculate' the max power... and some use completely nonsensical numbers.
For real power ratings, look at amps that use the CEA-2006 standards.
For years, people used an amp-clamp type meter, their speakers and a multimeter to calculate the power produced by their amp. The numbers were useless because the voltage and current were not in phase.
For many amps, the manufacturer uses the full peak voltage, no load to 'calculate' the max power... and some use completely nonsensical numbers.
For real power ratings, look at amps that use the CEA-2006 standards.
For years, people used an amp-clamp type meter, their speakers and a multimeter to calculate the power produced by their amp. The numbers were useless because the voltage and current were not in phase.
Hi Perry,
Have you or do you seriously test multiple amps?
Tomorrow, I'll send some interesting Soundstream waveforms and FFTs, at high and low power levels, with and without their pre-output-filter FB.
You can always use a Barry Gilbert-cell multiplier, etc to calculate output power. I've also carefully used log/antilog amps, but if their exponents don't match perfectly, they aren't good for audio. Mismatched exponents create harmonic distortion and then, you have to consider the effects of the harmonic terms.
I've also seen lots of stuff on the "perfection" of Class A amps. Not so, with single-ended amps. While they don't suffer crossover, they can be pretty nonlinear. Have to be overlapping push-pull to nominally minimize distortion.
Ron
Last edited:
I tested for maximum clean power output a long time ago but lost interest in that.
When I was repairing amps, I tested them to ensure that they were reliable and produced clean output to full power but didn't do any critical testing since many amps lie about their specs and wouldn't meet them, no matter how well they were repaired.
Things would have been different if I repaired high-end home equipment.
When I was repairing amps, I tested them to ensure that they were reliable and produced clean output to full power but didn't do any critical testing since many amps lie about their specs and wouldn't meet them, no matter how well they were repaired.
Things would have been different if I repaired high-end home equipment.
Gee, I've never owned or tested any high end equipment. The highest end for me was the Carver Sunfire at ~ $1,500 when it was intro'd. I really wanted a servo'd Velodyne but didn't want to spring for $2k. It was also a bit too large for our tiny condo's living room.
Around 1970, I DIY'd a 250 Watt all NPN power amp. 5 years later, I upped the power level to 2 kW, to drive a brushed 2 HP DC servo motor for a rotating CT gantry. My first day on the job, I saw some Mechanical Engineers struggling with a hydraulic servo. The floor of the room had puddles of hydraulic oil. I told them there had to be a cleaner way.
I was a college dropout USAF Veteran, back to school on the GI Bill, with only a slight idea what I'd bitten off - trying to stop about 1,000 lbs of rotating gantry melted the usual diodes across the output NPNs. Much more inertial energy than a 1.5 kW electromagnetic speaker array. Replacing the diodes was a quick fix. A nice design for 1977. I was so proud of my Candy Apple Red anodized heat sinks.
We sold 250 of those CTs at ~ $1.2 Million, each.
-----------------------------------------------------------
A few years later, evaluating several powered subs in a high end dealer's listening room, I connected my SONY Discman and test CD, at 16 Hz, to an 18 inch servo'd Velodyne and turned up the volume until I could hear some distortion. Then I turned down the volume slightly, until I didn't hear anything, with my ear near the speaker. The listening room windows were resonating. Should have had my sound meter or a scope with me, but I WAG'd the SPL at ~ 120 dB at 6 inches. I wonder what it did for my hearing.
That's the test - if you hear anything, it's distortion, not fundamental. Velodyne was ahead of the industry.
Around 1970, I DIY'd a 250 Watt all NPN power amp. 5 years later, I upped the power level to 2 kW, to drive a brushed 2 HP DC servo motor for a rotating CT gantry. My first day on the job, I saw some Mechanical Engineers struggling with a hydraulic servo. The floor of the room had puddles of hydraulic oil. I told them there had to be a cleaner way.
I was a college dropout USAF Veteran, back to school on the GI Bill, with only a slight idea what I'd bitten off - trying to stop about 1,000 lbs of rotating gantry melted the usual diodes across the output NPNs. Much more inertial energy than a 1.5 kW electromagnetic speaker array. Replacing the diodes was a quick fix. A nice design for 1977. I was so proud of my Candy Apple Red anodized heat sinks.
We sold 250 of those CTs at ~ $1.2 Million, each.
-----------------------------------------------------------
A few years later, evaluating several powered subs in a high end dealer's listening room, I connected my SONY Discman and test CD, at 16 Hz, to an 18 inch servo'd Velodyne and turned up the volume until I could hear some distortion. Then I turned down the volume slightly, until I didn't hear anything, with my ear near the speaker. The listening room windows were resonating. Should have had my sound meter or a scope with me, but I WAG'd the SPL at ~ 120 dB at 6 inches. I wonder what it did for my hearing.
That's the test - if you hear anything, it's distortion, not fundamental. Velodyne was ahead of the industry.
Maybe 15 years ago, on the front cover of an audiophile mag was what looked like an updated Fisher Ultralinear 50 Watt amplifier, the one with transformer primary taps, driving beampower tetrode screen grids. I almost choked with laughter over the $25k price. $50k for stereo.
Those same "audiophiles" probably paid $2k for 6 feet of Oxygenated Pure Copper 000 AWG welding cable speaker leads.
Those same "audiophiles" probably paid $2k for 6 feet of Oxygenated Pure Copper 000 AWG welding cable speaker leads.
So you're telling me that there's a bit of snake oil being pushed in the audio industry?... Get out of here. No way!.
In cars, a lot of what you hear is the resonance of all of the plastic and metal panels.
Low frequencies aren't very hard on your ears.
I'm not sure where it is but I think at about 90dB, human ears run out of xmax. They can't hear clearly above that. Using sealed headphones (using a high-pass crossover feeding them) in a bass-heavy environment makes a huge difference because your ears are not being hammered by the lows.
I thought the speaker wire was supposed to be Teflon jacketed, pure silver, stranded, supported on pedestals off of any surface and away from the other conductor with impedance compensation networks on the speaker end and if you couldn't hear the difference, you had a tin ear.
In cars, a lot of what you hear is the resonance of all of the plastic and metal panels.
Low frequencies aren't very hard on your ears.
I'm not sure where it is but I think at about 90dB, human ears run out of xmax. They can't hear clearly above that. Using sealed headphones (using a high-pass crossover feeding them) in a bass-heavy environment makes a huge difference because your ears are not being hammered by the lows.
I thought the speaker wire was supposed to be Teflon jacketed, pure silver, stranded, supported on pedestals off of any surface and away from the other conductor with impedance compensation networks on the speaker end and if you couldn't hear the difference, you had a tin ear.
Here are the Soundstream output Voltage and fft plots I'd mentioned yesterday. The plots are much more time consuming than comparable linear amplifier plots. To avoid alaising data can't be sampled at less than 220 kilosamples per second. Makes for lots of data points and I'm not into decimation or truncation.
1) The first RED plot is the 100 Hz output Voltage ~ 340 Watts and a representative input (BLU) waveform, set to nominally, match the output amplitude, so you can visually spot distortion. All the attached sine wave Voltages show a 400 usec start-up glitch, purposely left in place.
As you can see, the first RED output Voltage plot, without FB, shows a form of crossover distortion that results from digital shoot-through prevention. While it's not quite centered on zero, it performs and looks just like crossover distortion in a classic AB amp with insufficient crossover bias - for exactly the same reason.
2) Next is an fft of waveform 1) shows about 8.7% rms THD. Since the waveform and its distortion are near-symmetrical, the series is predominately odd harmonics.
3) The second RED plot is with 76 dB of SOUNDSTREAM-like, pre-LP Filter FB, with the input re-adjusted to provide an output level similar to 1). Very little visible output distortion. The plot is mislabeled. The power output is ~ 400 Watts, not 760 Watts.
4) Next is an FFT of waveform 3). At this high output level the rms THD is beautiful at < 0.12%
5) The third 100 Hz RED output plot is with the input reduced to produce a 5 Watt output. The plot erroneously shows 35 Watt. At the 5 Watt output level, without FB, distortion is quite high. I'm not sure yet, why the waveform is asymmetrical.
6) The fft of waveform 5) shows the THD is ~
Whenever I install any new gear, I sweep the house for resonances. Even China tea cups are Velcro'd down. There was a strange sound coming from our grandfather clock that I finally tracked down to the steel straps tying it to the studs - I've always done that to everything in quake country. The two steel straps were vibrating against the top of the clock. A couple of fuzzy Velcro straps stopped it.
I've swept ALL our cars and SUVs since the 90s. Our leased C-280 had a particularly irritating buzz. I removed the rear speaker sub-deck (remember, it was a leased car) and found two nylon push-in wire clamps weren't pushed in. There were two others simply lying on the deck. POS.
Many old car speakers, not shielded from the sun, developed bleached, torn cones or obnoxious voice coil rub.
If you want to know about piezoelectric, triboelectric and other subtle effects, there is an ancient book, Low Level Measurements by Keithley Instruments, who were routinely selling gear that measured femtoVolts and attoAmps before many of us even knew they had names. They suggested building your electronics in the air, like Bob Pease, on a sapphire or diamond substrate. Cost more than the US economy.
Call me cheap, but I've never used any more than 18 AWG zip cord for speaker cables. IF you need to go under carpets, spaced, wide copper straps work well. Folded-over heavy duty Aluminum foil also works.
On the subject of blown ears, I occasionally target shoot. Having previously tested pro headsets, I searched for ANC shooter's headsets. There is no such thing, at least for civilians. There are certainly lots of BS and deception. I tried a pair of SONY digitals on a B-25 flight. Prop noise overloaded them. Digitals aren't designed for impulse noise. I have no idea how well the pilot and co-pilot's analog David Clark's worked.
Concerned about noise-induced hearing loss (NIHL), I threw a paper together for a friend who volunteered as a range officer. I've recently posted it on a DIY Audio Tinnitus thread.
At one time I saw something like there were 10,000 speaker manufactures. There are loads of Elmer Gantry sales reps. Lock up your wife, partner and daughters. Buy them chastity belts and/or "Get Thee to a Nunnery."
Whenever I install any new gear, I sweep the house for resonances. Even China tea cups are Velcro'd down. There was a strange sound coming from our grandfather clock that I finally tracked down to the steel straps tying it to the studs - I've always done that to everything in quake country. The two steel straps were vibrating against the top of the clock. A couple of fuzzy Velcro straps stopped it.
I've swept ALL our cars and SUVs since the 90s. Our leased C-280 had a particularly irritating buzz. I removed the rear speaker sub-deck (remember, it was a leased car) and found two nylon push-in wire clamps weren't pushed in. There were two others simply lying on the deck. POS.
Many old car speakers, not shielded from the sun, developed bleached, torn cones or obnoxious voice coil rub.
If you want to know about piezoelectric, triboelectric and other subtle effects, there is an ancient book, Low Level Measurements by Keithley Instruments, who were routinely selling gear that measured femtoVolts and attoAmps before many of us even knew they had names. They suggested building your electronics in the air, like Bob Pease, on a sapphire or diamond substrate. Cost more than the US economy.
Call me cheap, but I've never used any more than 18 AWG zip cord for speaker cables. IF you need to go under carpets, spaced, wide copper straps work well. Folded-over heavy duty Aluminum foil also works.
On the subject of blown ears, I occasionally target shoot. Having previously tested pro headsets, I searched for ANC shooter's headsets. There is no such thing, at least for civilians. There are certainly lots of BS and deception. I tried a pair of SONY digitals on a B-25 flight. Prop noise overloaded them. Digitals aren't designed for impulse noise. I have no idea how well the pilot and co-pilot's analog David Clark's worked.
Concerned about noise-induced hearing loss (NIHL), I threw a paper together for a friend who volunteered as a range officer. I've recently posted it on a DIY Audio Tinnitus thread to jaxboy.
1) The first RED plot is the 100 Hz output Voltage ~ 340 Watts and a representative input (BLU) waveform, set to nominally, match the output amplitude, so you can visually spot distortion. All the attached sine wave Voltages show a 400 usec start-up glitch, purposely left in place.
As you can see, the first RED output Voltage plot, without FB, shows a form of crossover distortion that results from digital shoot-through prevention. While it's not quite centered on zero, it performs and looks just like crossover distortion in a classic AB amp with insufficient crossover bias - for exactly the same reason.
2) Next is an fft of waveform 1) shows about 8.7% rms THD. Since the waveform and its distortion are near-symmetrical, the series is predominately odd harmonics.
3) The second RED plot is with 76 dB of SOUNDSTREAM-like, pre-LP Filter FB, with the input re-adjusted to provide an output level similar to 1). Very little visible output distortion. The plot is mislabeled. The power output is ~ 400 Watts, not 760 Watts.
4) Next is an FFT of waveform 3). At this high output level the rms THD is beautiful at < 0.12%
5) The third 100 Hz RED output plot is with the input reduced to produce a 5 Watt output. The plot erroneously shows 35 Watt. At the 5 Watt output level, without FB, distortion is quite high. I'm not sure yet, why the waveform is asymmetrical.
6) The fft of waveform 5) shows the THD is ~
At one time I saw something like there were 10,000 speaker manufactures. There are loads of Elmer Gantry sales reps. Lock up your wife, partner and daughters. Buy them chastity belts and/or "Get Thee to a Nunnery."
Whenever I install any new gear, I sweep the house for resonances. Even China tea cups are Velcro'd down. There was a strange sound coming from our grandfather clock that I finally tracked down to the steel straps tying it to the studs - I've always done that to everything in quake country. The two steel straps were vibrating against the top of the clock. A couple of fuzzy Velcro straps stopped it.
I've swept ALL our cars and SUVs since the 90s. Our leased C-280 had a particularly irritating buzz. I removed the rear speaker sub-deck (remember, it was a leased car) and found two nylon push-in wire clamps weren't pushed in. There were two others simply lying on the deck. POS.
Many old car speakers, not shielded from the sun, developed bleached, torn cones or obnoxious voice coil rub.
If you want to know about piezoelectric, triboelectric and other subtle effects, there is an ancient book, Low Level Measurements by Keithley Instruments, who were routinely selling gear that measured femtoVolts and attoAmps before many of us even knew they had names. They suggested building your electronics in the air, like Bob Pease, on a sapphire or diamond substrate. Cost more than the US economy.
Call me cheap, but I've never used any more than 18 AWG zip cord for speaker cables. IF you need to go under carpets, spaced, wide copper straps work well. Folded-over heavy duty Aluminum foil also works.
On the subject of blown ears, I occasionally target shoot. Having previously tested pro headsets, I searched for ANC shooter's headsets. There is no such thing, at least for civilians. There are certainly lots of BS and deception. I tried a pair of SONY digitals on a B-25 flight. Prop noise overloaded them. Digitals aren't designed for impulse noise. I have no idea how well the pilot and co-pilot's analog David Clark's worked.
Concerned about noise-induced hearing loss (NIHL), I threw a paper together for a friend who volunteered as a range officer. I've recently posted it on a DIY Audio Tinnitus thread.
At one time I saw something like there were 10,000 speaker manufactures. There are loads of Elmer Gantry sales reps. Lock up your wife, partner and daughters. Buy them chastity belts and/or "Get Thee to a Nunnery."
Whenever I install any new gear, I sweep the house for resonances. Even China tea cups are Velcro'd down. There was a strange sound coming from our grandfather clock that I finally tracked down to the steel straps tying it to the studs - I've always done that to everything in quake country. The two steel straps were vibrating against the top of the clock. A couple of fuzzy Velcro straps stopped it.
I've swept ALL our cars and SUVs since the 90s. Our leased C-280 had a particularly irritating buzz. I removed the rear speaker sub-deck (remember, it was a leased car) and found two nylon push-in wire clamps weren't pushed in. There were two others simply lying on the deck. POS.
Many old car speakers, not shielded from the sun, developed bleached, torn cones or obnoxious voice coil rub.
If you want to know about piezoelectric, triboelectric and other subtle effects, there is an ancient book, Low Level Measurements by Keithley Instruments, who were routinely selling gear that measured femtoVolts and attoAmps before many of us even knew they had names. They suggested building your electronics in the air, like Bob Pease, on a sapphire or diamond substrate. Cost more than the US economy.
Call me cheap, but I've never used any more than 18 AWG zip cord for speaker cables. IF you need to go under carpets, spaced, wide copper straps work well. Folded-over heavy duty Aluminum foil also works.
On the subject of blown ears, I occasionally target shoot. Having previously tested pro headsets, I searched for ANC shooter's headsets. There is no such thing, at least for civilians. There are certainly lots of BS and deception. I tried a pair of SONY digitals on a B-25 flight. Prop noise overloaded them. Digitals aren't designed for impulse noise. I have no idea how well the pilot and co-pilot's analog David Clark's worked.
Concerned about noise-induced hearing loss (NIHL), I threw a paper together for a friend who volunteered as a range officer. I've recently posted it on a DIY Audio Tinnitus thread to jaxboy.
Hi Perry,
Here are the real Soundstream output Voltage and fft plots. They're more time consuming than comparable linear amplifier plots. To avoid alaising, data can't be sampled at less than 220 kilosamples per second. Makes for lots of data points and calculation time.
1) The first RED plot is the 100 Hz output Voltage ~ 340 Watts and a representative input (BLU) waveform, set to nominally, match the output, so you can visually spot distortion. All the attached sine waves show a 400 usec start-up glitch, purposely left in place.
The first RED output Voltage plot, without FB, shows a form of crossover distortion that results from digital shoot-through prevention. While it's not quite centered on zero, it performs and looks just like crossover distortion in a classic AB amp with insufficient crossover bias.
2) An fft of waveform 1) shows the rms THD without FB is about 8.7%. Since the waveform and its distortion are near-symmetrical, the series is predominately odd harmonics.
-----------------------------------------------------------------------------
3) The second RED output Voltage plot is with 76 dB of Soundstream-like, pre-LP Filter FB, with the input re-adjusted to provide an output level similar to 1). Very little visible output distortion. The power output is mislabeled. It's ~ 400 Watts, not 760.
4) An fft of waveform 3) shows the rms THD with FB is excellent, at < 0.12%
----------------------------------------------------------------------------
5) The third 100 Hz RED output Voltage plot is with the input reduced, to produce a 5 Watt output. The plot erroneously says 35 Watts. At the 5 Watt output level, without FB, distortion is quite high, as expected. I'm not sure yet, exactly why the waveform is asymmetrical.
6) An fft of waveform 5) shows the rms THD with no FB is 13.4%
-----------------------------------------------------------------------------
7) The fourth 100 Hz RED output plot is a 5 Watt output with 76 dB of FB. It's mislabeled. and should say With FB. Visually, it looks remarkably good. I was expecting much worse, even with FB.
8) An fft of waveform 7) shows the rms THD with FB is ~ 0.1%. That comes as a complete surprise.
-----------------------------------------------------------------------------
As you can see, the percentage rms THD, without FB, increases as the output level decreases. What that says is the bad crossover stuff, clustered around the zero crossings, remains ~ constant as the good stuff decreases. That's entirely expected.
With FB, I'd expected much of the same. It didn't work that way, and I've yet to understand why.
----------------------------------------------------------------------------
I looked at the both the FET Emitter follower Gate drives and the FET Drain currents at 5 Watts on a pulse-by-pulse basis, to see if there's any FB-induced, low-level shoot-through, as the output crosses 0V. On first pass, I couldn't find any.
Soundstream's FB seems to work quite well, even down to low power, low listening levels.
Here are the real Soundstream output Voltage and fft plots. They're more time consuming than comparable linear amplifier plots. To avoid alaising, data can't be sampled at less than 220 kilosamples per second. Makes for lots of data points and calculation time.
1) The first RED plot is the 100 Hz output Voltage ~ 340 Watts and a representative input (BLU) waveform, set to nominally, match the output, so you can visually spot distortion. All the attached sine waves show a 400 usec start-up glitch, purposely left in place.
The first RED output Voltage plot, without FB, shows a form of crossover distortion that results from digital shoot-through prevention. While it's not quite centered on zero, it performs and looks just like crossover distortion in a classic AB amp with insufficient crossover bias.
2) An fft of waveform 1) shows the rms THD without FB is about 8.7%. Since the waveform and its distortion are near-symmetrical, the series is predominately odd harmonics.
-----------------------------------------------------------------------------
3) The second RED output Voltage plot is with 76 dB of Soundstream-like, pre-LP Filter FB, with the input re-adjusted to provide an output level similar to 1). Very little visible output distortion. The power output is mislabeled. It's ~ 400 Watts, not 760.
4) An fft of waveform 3) shows the rms THD with FB is excellent, at < 0.12%
----------------------------------------------------------------------------
5) The third 100 Hz RED output Voltage plot is with the input reduced, to produce a 5 Watt output. The plot erroneously says 35 Watts. At the 5 Watt output level, without FB, distortion is quite high, as expected. I'm not sure yet, exactly why the waveform is asymmetrical.
6) An fft of waveform 5) shows the rms THD with no FB is 13.4%
-----------------------------------------------------------------------------
7) The fourth 100 Hz RED output plot is a 5 Watt output with 76 dB of FB. It's mislabeled. and should say With FB. Visually, it looks remarkably good. I was expecting much worse, even with FB.
8) An fft of waveform 7) shows the rms THD with FB is ~ 0.1%. That comes as a complete surprise.
-----------------------------------------------------------------------------
As you can see, the percentage rms THD, without FB, increases as the output level decreases. What that says is the bad crossover stuff, clustered around the zero crossings, remains ~ constant as the good stuff decreases. That's entirely expected.
With FB, I'd expected much of the same. It didn't work that way, and I've yet to understand why.
----------------------------------------------------------------------------
I looked at the both the FET Emitter follower Gate drives and the FET Drain currents at 5 Watts on a pulse-by-pulse basis, to see if there's any FB-induced, low-level shoot-through, as the output crosses 0V. On first pass, I couldn't find any.
Soundstream's FB seems to work quite well, even down to low power, low listening levels.
I'm rashly assuming they're identical or at least similar.
SS's taking the Voltage FB pre-filter is the key to solving Class D crossover distortion. All my FB was pre-filter.
Once I evaluate its thermal and supply Voltage drift, I'll try to use the Soundstream as a DC servo amp. That was the reason I needed to know what was inside. I couldn't find a real 12V DC servo amp, at least one that didn't break the bank
I think I'd mentioned buying several of those cheapo ($22), 600 Watt Class D's, but their companion cheapo ($12) power modules didn't have negative outputs. That was when, two months ago, I decided to take my chances with a complete amplifier. Since I knew zip about high power consumer Class Ds, my choice was based on this almost comical YouTube face-off of its bigger brother, the AR1.8000D:
A few days back, I stumbled on someone saying the Class D crossover solution was creating narrow, alternating polarity pulses as the output approaches and leaves zero crossings, but that process has to be controlled, otherwise it's noise or dither.
If it's done, well outside the listening BW, you don't hear (much of) the noise or dither. If the dither is pseudo-random and predictable, in theory, it can be fully removed, post-process. That's what's done in modern Delta Sigma ADCs and DACs. Since I'm not fully conversant in iterative or recursive filters and Z Transforms, I still stumble on those chapters and DSP in general.
I was using my loaner Norseworthy, Schreier and Temes' Delta-Sigma Data Converters to prop up my camera's copy stand. I slid it out, yesterday, to look at the chapters on Dither. I'd give it back to the owner, but he died 20 years ago.
Now, all I need is a no BS explanation of how pre-filter Class D FB actually works.
SS's taking the Voltage FB pre-filter is the key to solving Class D crossover distortion. All my FB was pre-filter.
Once I evaluate its thermal and supply Voltage drift, I'll try to use the Soundstream as a DC servo amp. That was the reason I needed to know what was inside. I couldn't find a real 12V DC servo amp, at least one that didn't break the bank
I think I'd mentioned buying several of those cheapo ($22), 600 Watt Class D's, but their companion cheapo ($12) power modules didn't have negative outputs. That was when, two months ago, I decided to take my chances with a complete amplifier. Since I knew zip about high power consumer Class Ds, my choice was based on this almost comical YouTube face-off of its bigger brother, the AR1.8000D:
A few days back, I stumbled on someone saying the Class D crossover solution was creating narrow, alternating polarity pulses as the output approaches and leaves zero crossings, but that process has to be controlled, otherwise it's noise or dither.
If it's done, well outside the listening BW, you don't hear (much of) the noise or dither. If the dither is pseudo-random and predictable, in theory, it can be fully removed, post-process. That's what's done in modern Delta Sigma ADCs and DACs. Since I'm not fully conversant in iterative or recursive filters and Z Transforms, I still stumble on those chapters and DSP in general.
I was using my loaner Norseworthy, Schreier and Temes' Delta-Sigma Data Converters to prop up my camera's copy stand. I slid it out, yesterday, to look at the chapters on Dither. I'd give it back to the owner, but he died 20 years ago.
Now, all I need is a no BS explanation of how pre-filter Class D FB actually works.
Before the filter, I don't think there is much of a difference in the way the feedback works, compared to anything else. If there is any distortion, the circuit sends a correction as strong as needed to clean up the signal. After the filter, the circuit can be significantly different.
The following amp has minimal filtering in the feedback circuit. Look at PDF pages 29 and 30.
http://www.bcae1.com/temp/infinity - REF611asm.pdf
Can you view the audio signal into the driver board with and without a load? That should show how the feedback circuit is compensating.
The following amp has minimal filtering in the feedback circuit. Look at PDF pages 29 and 30.
http://www.bcae1.com/temp/infinity - REF611asm.pdf
Can you view the audio signal into the driver board with and without a load? That should show how the feedback circuit is compensating.
Thanks for the Infinity schematic.
I'm a huge feedback fan.
It looks like the Infinity is infinitely more complex than the Soundstream. It uses a full H bridge, with pre-filter DC FB from both sides of the H to a diff amp driving the modulator. The Infinity's EEs didn't seem driven by one my all-time heroes, self-taught EE, Madman Earl Muntz. Earl would look over the shoulders of his designers and ask "What does that component do?" If a designer couldn't answer to Earl's satisfaction, he'd say "Get rid of it.". It's the driving force behind every rabid designer, especially analog. Another splendid example is Li-Chen Wang, who designed Palo Alto Tiny Basic, with less then 2kB of memory.
https://en.wikipedia.org/wiki/Madman_Muntz
https://en.wikipedia.org/wiki/Li-Chen_Wang#Palo_Alto_Tiny_BASIC
------------------------------------------------------------------------------------
Without a lead term, taking FB from the output side of a Class D's second order RLC output filter is an open invitation to instability. Using pre-filter FB is a lot more stable if you can keep the FB carrier from overloading the stage that sums the FB with the audio, just before the modulator. I tweaked the existing single pole FB filter cap, so the p-p carrier at the output of that summer idles at about 0.5V p-p.
The Soundstream's FB path to the summer input is DC,. A 10 uF coupling cap between the summer output and the modulator input breaks that DC path, so the FB loop has no control over output DC drift. Shorting that 10 uF works just fine.
My concern over carrier in the overall FB loop is unsubstantiated. You're absolutely right. With a small amount of carrier, it works just like an ordinary FB loop. The are simple ways to synchronously remove the carrier, but it doesn't look like it's worth the effort.
In the old days, when tube stuff had poor DC drift, but reasonably stable AC gain, they'd chop the input to make it pulsating DC (essentially AC), amplify it, then synchronously demodulate and filter it to recoup an amplified version of the original DC input. Modern ultra-low offset, ultra low drift op amps still use synchronous mod/demods, but people don't use them in low noise apps.
I haven't put everything back together yet, for overall testing. Maybe in two or three days.
-----------------------------------------------------------------------------------
Remember what I said about thinking Class D percent crossover THD would increase at at low listening levels? I've looked at several data sheets, including the darling TI TPA3250. It's right there on page 1 of the attached data sheet.
Approx [1/(Output power)], below 1 Watt. So far, I haven't seen it Soundstream sims. Maybe once I put it together?
-------------------------------------------------------------------
While I was looking for the TPA3250 Class D amplifier data sheet, I stumbled on a chip and its EVAL board that I thought TI had stopped selling 10 years ago. Before TI acquired National Semiconductor, I'd designed and patented the CW Doppler section of the AFE5812 Ultrasound Analog Front End chip. Another EE and I designed and tested the chip's EVAL board. Our best CW Doppler tester turned out to be the Audio Precision 2700 we lifted from National's Audio Group. I wish I had one now, but I'm too cheap to buy it.
Besides, many of today's sound boards and even freebie fft s/w can almost meet the obsolete 2700's performance.
If I've confused you, the MRI stuff preceded the ultrasound stuff by 25 years.
Jack of All Trades and Master of None,
I'm a huge feedback fan.
It looks like the Infinity is infinitely more complex than the Soundstream. It uses a full H bridge, with pre-filter DC FB from both sides of the H to a diff amp driving the modulator. The Infinity's EEs didn't seem driven by one my all-time heroes, self-taught EE, Madman Earl Muntz. Earl would look over the shoulders of his designers and ask "What does that component do?" If a designer couldn't answer to Earl's satisfaction, he'd say "Get rid of it.". It's the driving force behind every rabid designer, especially analog. Another splendid example is Li-Chen Wang, who designed Palo Alto Tiny Basic, with less then 2kB of memory.
https://en.wikipedia.org/wiki/Madman_Muntz
https://en.wikipedia.org/wiki/Li-Chen_Wang#Palo_Alto_Tiny_BASIC
------------------------------------------------------------------------------------
Without a lead term, taking FB from the output side of a Class D's second order RLC output filter is an open invitation to instability. Using pre-filter FB is a lot more stable if you can keep the FB carrier from overloading the stage that sums the FB with the audio, just before the modulator. I tweaked the existing single pole FB filter cap, so the p-p carrier at the output of that summer idles at about 0.5V p-p.
The Soundstream's FB path to the summer input is DC,. A 10 uF coupling cap between the summer output and the modulator input breaks that DC path, so the FB loop has no control over output DC drift. Shorting that 10 uF works just fine.
My concern over carrier in the overall FB loop is unsubstantiated. You're absolutely right. With a small amount of carrier, it works just like an ordinary FB loop. The are simple ways to synchronously remove the carrier, but it doesn't look like it's worth the effort.
In the old days, when tube stuff had poor DC drift, but reasonably stable AC gain, they'd chop the input to make it pulsating DC (essentially AC), amplify it, then synchronously demodulate and filter it to recoup an amplified version of the original DC input. Modern ultra-low offset, ultra low drift op amps still use synchronous mod/demods, but people don't use them in low noise apps.
I haven't put everything back together yet, for overall testing. Maybe in two or three days.
-----------------------------------------------------------------------------------
Remember what I said about thinking Class D percent crossover THD would increase at at low listening levels? I've looked at several data sheets, including the darling TI TPA3250. It's right there on page 1 of the attached data sheet.
Approx [1/(Output power)], below 1 Watt. So far, I haven't seen it Soundstream sims. Maybe once I put it together?
-------------------------------------------------------------------
While I was looking for the TPA3250 Class D amplifier data sheet, I stumbled on a chip and its EVAL board that I thought TI had stopped selling 10 years ago. Before TI acquired National Semiconductor, I'd designed and patented the CW Doppler section of the AFE5812 Ultrasound Analog Front End chip. Another EE and I designed and tested the chip's EVAL board. Our best CW Doppler tester turned out to be the Audio Precision 2700 we lifted from National's Audio Group. I wish I had one now, but I'm too cheap to buy it.
Besides, many of today's sound boards and even freebie fft s/w can almost meet the obsolete 2700's performance.
If I've confused you, the MRI stuff preceded the ultrasound stuff by 25 years.
Jack of All Trades and Master of None,
Last edited:
Hi Perry,
I've been inactive for almost a week with other issues, the latest was a failing cable
modem/router.
While doing a few more SS sims, trying to determine the amount of inverse pre-filter FB that would be stable and how to generally measure phase/gain margin in a Class D, I started looking into the 3 kHz ringing that's appeared in all my SS gain plots and FFTs.
The 120 uH inductor and 22 uF I'd been using were from the schematic you'd provided. I've yet to measure the values in my own amplifier. When I did sims based on the provided schematic, I'd added 10 milliOhms of inductor series resistance, from the very beginning. The value was just chosen, based on tolerable IR drop.
Changing R and the output capacitor can effect Q, but they have other implications. I'd never dealt with filter ringing, because I had no real experience with Class D amplifiers or their output filters
The 3 kHz ringing is strictly set by the 120 uH and 22 uF output cap. The Q is ~ 3 and can be reduced by increasing my 10 milliOhms series resistance. but that can't go far, since it harms efficiency. I looked in diyAudio for ringing solutions but didn't find any - not to say they don't exist - I just didn't find any.
So I tried to sum some post-filter FB along with my pre-filter FB. I decreased my pre-filter FB to half and applied an equal amount of post-filter FB. I ended up with slowly decaying ringing that approached oscillation. So I increased my pre-filter FB to ~ 75% of its original value and decreased the post-filter FB to ~ 25% and almost all evidence of ringing went away. Phase alignment should improve performance.
What that says is a proper amplitude-and-phased amounts of both pre-filter and post-filter FB can reduce distortion, improve dynamics, frequency response and reduce output filter Q-related ringing.
What I've found belongs in a general Class D stream, not just in car Class D. I'll try some time in the pre-post pro-con streams.
Ron
Last edited:
I found this. It takes the feedback after the inductor and uses a slightly different circuit.
https://elektrotanya.com/showresult?what=EARTHQUAKE MINI D2000&kategoria=&kat2=all
https://elektrotanya.com/showresult?what=EARTHQUAKE MINI D2000&kategoria=&kat2=all
Hi Perry,
Thanks for the link to the MINI D2000.
Here's what I've found so far:
1) The D2000's FB is 100% post-output-filter.
2) The SLA-marked schematic you'd sent, and my own AR1.4500D, both use 100% pre-filter FB. I've traced the entire feedback loop for the 4500D.
3) Both the SLA and 4500D use a summing op amp before the PWM. The inverting (-) input of the summer can accept either SOUNDSTREAM"s pre-filter or post-filter FB. In both SOUNDSTREAM amplifiers, audio is fed to the non-inverting (+) op amp input. It could just as easily feed the inverting input.
4) Using All Pre-filter FB:
If the FB is all pre-filter, the intent is to apply as much FB as possible to the summer, without a) instability or b) significant reduction of the summer's dynamic range with excess 100 kHz carrier in the FB signal. I've arbitrarily limited carrier swing at the summer's output to 1V peak-to-peak.
At least 30 dB of pre-filter FB can be applied to the summer's inverting input without encountering a) or b). 30 dB is enough FB to reduce crossover distortion at full output from ~ 3% to ~ 0.1%,
Ignoring the small crossover gap, at a 50% Duty Cycle, the unloaded, unfiltered, +/-60V amplifier max output contains 54 Vrms at the 100 kHz fundamental, 18 Vrms at the 300 kHz 3rd harmonic, 10.8 Vrms at the 500 kHz 5th harmonic, 7.7 Vrms at the 700 kHz 7th harmonic, etc..
To attenuate FB carrier at least 60 dB at 100 kHz, there's a dominant, single pole LP at 100 Hz. Above 100 Hz, carrier is attenuated 20 dB per decade, or 60 dB over 3 decades (100 Hz to 100 kHz). That single 100 Hz dominant LP pole makes the pre-filter FB loop sluggish.
In my simulations, I've removed the cap across the summer's FB resistor and the two 100 nF caps shunting the summer's output and PWM's input. What I've simulated in the SOUNDSTREAM is a 2 pole FB roll-off starting at 3.1 kHz, and converting to a single pole at 100 kHz and above. That 2 pole/1 pole arrangement provides the same amount of 100 kHz FB carrier attenuation, while allowing 6 times the "flat" audio BW.
Assume the pre-filter FB loop is actually prefect. There's the issue of what the 2nd order RLC output filter does. Its main purpose is to keep the 100 kHz from wasting power by heating the speaker. It's a very rare occasion that 100kHz can be heard by a human or creates an audible artifact.
The SLA filter's 120 uH and 22uF, driving a 2 Ohm speaker, has a resonant frequency of 3.1 kHz, and a Q ~ 0.86, not 3 as I'd previously stated. For a 2 Ohm load, the damping factor or Zeta is 1/(2Q) or 0.58. The Q is lower with a 1 Ohm load and higher, with a 4 Ohm load. For minimum overshoot and minimum settling time, both Q and Zeta can be simultaneously optimized at .707
I haven't yet measured the 4500's output L or C.
With 2 Ohms on the SLA's output, there is a ~ 1/2 dB gain peak at 3.1 kHz and a small amount of ringing, decaying down to 1/e or 37% in 1 cycle, down to 1/(e squared) or 14% in 2 cycles and down to 1/(e cubed) or 5% in 3 cycles.......
5) Using All Post-filter FB:
That same SLA output filter has ~ 5 degrees of phase shift at 310 Hz, 90 degrees at 3.1 kHz and 175 degrees at 31 kHz. Without any other lead, lag or lead/lag FB, the output filter's phase shift will be 135 degrees at 1 kHz and the amplifier will start to peak as the phase margin drops below (180-135) or 45 degrees, limiting the amount of all-post-filter FB that can be applied.
6) Combining Pre and Post-filter FB:
Initially ignore the output filter, and apply ~ 30 dB of pre-filter FB to optimize the pre-filter loop's frequency and dynamic response. Then apply about 2/3 that same FB level, post-filter, to damp output filter ringing. The pre-filter loop acts like feed-forward FB to partially compensate post-filter phase lag and ringing. The pre-post FB combo works very nicely.
-----------------------------------------------------------------------------
I'm ready to implement the combined pre-post FB in hardware, then test the amplifier.
While I don't need it, I'm looking at ways to make the output FETs look like a low power Class AB below ~ 10 Watts output, minimizing the Class D's low power crossover distortion.
Ron
Thanks for the link to the MINI D2000.
Here's what I've found so far:
1) The D2000's FB is 100% post-output-filter.
2) The SLA-marked schematic you'd sent, and my own AR1.4500D, both use 100% pre-filter FB. I've traced the entire feedback loop for the 4500D.
3) Both the SLA and 4500D use a summing op amp before the PWM. The inverting (-) input of the summer can accept either SOUNDSTREAM"s pre-filter or post-filter FB. In both SOUNDSTREAM amplifiers, audio is fed to the non-inverting (+) op amp input. It could just as easily feed the inverting input.
4) Using All Pre-filter FB:
If the FB is all pre-filter, the intent is to apply as much FB as possible to the summer, without a) instability or b) significant reduction of the summer's dynamic range with excess 100 kHz carrier in the FB signal. I've arbitrarily limited carrier swing at the summer's output to 1V peak-to-peak.
At least 30 dB of pre-filter FB can be applied to the summer's inverting input without encountering a) or b). 30 dB is enough FB to reduce crossover distortion at full output from ~ 3% to ~ 0.1%,
Ignoring the small crossover gap, at a 50% Duty Cycle, the unloaded, unfiltered, +/-60V amplifier max output contains 54 Vrms at the 100 kHz fundamental, 18 Vrms at the 300 kHz 3rd harmonic, 10.8 Vrms at the 500 kHz 5th harmonic, 7.7 Vrms at the 700 kHz 7th harmonic, etc..
To attenuate FB carrier at least 60 dB at 100 kHz, there's a dominant, single pole LP at 100 Hz. Above 100 Hz, carrier is attenuated 20 dB per decade, or 60 dB over 3 decades (100 Hz to 100 kHz). That single 100 Hz dominant LP pole makes the pre-filter FB loop sluggish.
In my simulations, I've removed the cap across the summer's FB resistor and the two 100 nF caps shunting the summer's output and PWM's input. What I've simulated in the SOUNDSTREAM is a 2 pole FB roll-off starting at 3.1 kHz, and converting to a single pole at 100 kHz and above. That 2 pole/1 pole arrangement provides the same amount of 100 kHz FB carrier attenuation, while allowing 6 times the "flat" audio BW.
Assume the pre-filter FB loop is actually prefect. There's the issue of what the 2nd order RLC output filter does. Its main purpose is to keep the 100 kHz from wasting power by heating the speaker. It's a very rare occasion that 100kHz can be heard by a human or creates an audible artifact.
The SLA filter's 120 uH and 22uF, driving a 2 Ohm speaker, has a resonant frequency of 3.1 kHz, and a Q ~ 0.86, not 3 as I'd previously stated. For a 2 Ohm load, the damping factor or Zeta is 1/(2Q) or 0.58. The Q is lower with a 1 Ohm load and higher, with a 4 Ohm load. For minimum overshoot and minimum settling time, both Q and Zeta can be simultaneously optimized at .707
I haven't yet measured the 4500's output L or C.
With 2 Ohms on the SLA's output, there is a ~ 1/2 dB gain peak at 3.1 kHz and a small amount of ringing, decaying down to 1/e or 37% in 1 cycle, down to 1/(e squared) or 14% in 2 cycles and down to 1/(e cubed) or 5% in 3 cycles.......
5) Using All Post-filter FB:
That same SLA output filter has ~ 5 degrees of phase shift at 310 Hz, 90 degrees at 3.1 kHz and 175 degrees at 31 kHz. Without any other lead, lag or lead/lag FB, the output filter's phase shift will be 135 degrees at 1 kHz and the amplifier will start to peak as the phase margin drops below (180-135) or 45 degrees, limiting the amount of all-post-filter FB that can be applied.
6) Combining Pre and Post-filter FB:
Initially ignore the output filter, and apply ~ 30 dB of pre-filter FB to optimize the pre-filter loop's frequency and dynamic response. Then apply about 2/3 that same FB level, post-filter, to damp output filter ringing. The pre-filter loop acts like feed-forward FB to partially compensate post-filter phase lag and ringing. The pre-post FB combo works very nicely.
-----------------------------------------------------------------------------
I'm ready to implement the combined pre-post FB in hardware, then test the amplifier.
While I don't need it, I'm looking at ways to make the output FETs look like a low power Class AB below ~ 10 Watts output, minimizing the Class D's low power crossover distortion.
Ron
Last edited:
After several month's absence for family health issues, but with a few more chances to LTSpice model SoundStream's DIY Class D modulator, I've concluded their entire signal chain, including the dead time(s) to prevent potentially destructive output cross-currents, DOES NOT produce crossover distortion. The output stage is just as linear, with the same gain at crossover as everywhere else.
Everything I'd previously told you about Class D crossover distortion was total BS.
I'd mentioned previously, a nominal 50/50 mix of pre and post LC filter FB optimizes dynamic performance and frequency response. Use of total pre-filter FB leaves the output Damping Factor, or its inverse, Q, at the mercy of the speaker (driver) impedance. My 70 uH, 22 uF LC is significantly underdamped if the speaker impedance exceeds 2 Ohms.
Use of total post filter FB invariably leads to HF phase shift, ringing and potential oscillation. Any use of stabilizing resistance in parallel with the output inductor or in series with the output capacitor, produces serious ripple and negative efficiency effects. If you try to use post-output-filter Lead-Lag FB to correct HF phase shift, it feeds too much 110 kHz ripple back to the modulator, making it switch at 220, 330 or 440 kHz, killing efficiency.
I've used 30 dB of combined 50% pre and 50% post FB to produce a slightly underdamped 20% overshoot at the 4 kHz resonant frequency of the 70 uH, 22 uF output filter. The dynamic response with a 500 Hz square wave input is quite acceptable. Overshoot can be minimized with a single pole audio input RC Low Pass.
In process of modding my Soundstream, I smoked 4 output FETs. While I wasn't paying attention, the 60VDC on my $300 subwoofer's voice coil made it quite odoriferous. A minor $450 mistake.
At $1.25 each, I have a dozen new output FETs on hand. But, just in case, a new Soundstream and subwoofer are sitting in my garage lab, unboxed.
After all this Class D stuff, I'm starting to look, lovingly, at my 6 year old Crunch Class AB PZX 1200.4. While I was initially put off by it's numerous output discretes, I think modding it as a DC servo would have been a lot less of a pain in the butt.
Last edited: