lol!
Hi Bob. I have your latest book but haven't got very far into it. Maybe we are of similar age, I started with tubes at a tender age. That included my first oscilloscope (Stark, 500 KHz single trace, no trigger).
HP made wonderfully accurate meters, the analogue kind! I finally have a 3400A. These are useful even though I have 34401A and 34465A meters. I used a 400EL earlier in time at my shop. That stayed when I sold the shop (too bad). If I didn't get the RTX, the QA403 would certainly be on my bench. Cheap, a real bargain.
I also have a couple HP 339A, a few 334A (always wanted a 8903B) and a ShibaSoku AD725C (from David Barber, RIP). I still use those as well. Even the old 3581A has uses, I have a couple. Their noise floor was really the only problem they had. Modern approaches could revitalize some of them. A THD meter is extremely valuable for seeing actual crossover distortion on a scope. Another tool employed on the bench. Never discount those older instruments!
In the practical world, I have seen my share of DC servos designed without understanding how they can affect things. Hence my comments. To be honest, I don't like DC servos that much. Yes, they can work fine, but they add complexity and cost to a design. Some of our designs before the servo became popular (DuoBeta anyone?) worked just fine and the only distortion you could be worried about was LF distortion numbers. To be honest, loudspeakers are the limiting factor there. I do agree we should strive to get distortion as low as reasonably possible unless we give up something else more valuable.
Thank you in advance for examining the topic of DC servos.
We agree completely on design objectives and understanding test measurements. Something doesn't make sense, go search for answers. Finding the oscillation was good sleuthing. This (bursts of oscillation) isn't that uncommon even today. Especially when equipment has been serviced.
Hi Bob. I have your latest book but haven't got very far into it. Maybe we are of similar age, I started with tubes at a tender age. That included my first oscilloscope (Stark, 500 KHz single trace, no trigger).
HP made wonderfully accurate meters, the analogue kind! I finally have a 3400A. These are useful even though I have 34401A and 34465A meters. I used a 400EL earlier in time at my shop. That stayed when I sold the shop (too bad). If I didn't get the RTX, the QA403 would certainly be on my bench. Cheap, a real bargain.
I also have a couple HP 339A, a few 334A (always wanted a 8903B) and a ShibaSoku AD725C (from David Barber, RIP). I still use those as well. Even the old 3581A has uses, I have a couple. Their noise floor was really the only problem they had. Modern approaches could revitalize some of them. A THD meter is extremely valuable for seeing actual crossover distortion on a scope. Another tool employed on the bench. Never discount those older instruments!
In the practical world, I have seen my share of DC servos designed without understanding how they can affect things. Hence my comments. To be honest, I don't like DC servos that much. Yes, they can work fine, but they add complexity and cost to a design. Some of our designs before the servo became popular (DuoBeta anyone?) worked just fine and the only distortion you could be worried about was LF distortion numbers. To be honest, loudspeakers are the limiting factor there. I do agree we should strive to get distortion as low as reasonably possible unless we give up something else more valuable.
Thank you in advance for examining the topic of DC servos.
We agree completely on design objectives and understanding test measurements. Something doesn't make sense, go search for answers. Finding the oscillation was good sleuthing. This (bursts of oscillation) isn't that uncommon even today. Especially when equipment has been serviced.
Hi Bob et al,
I've followed your tutorial on the QA403 with interest. Would you expound on the relative capabilities to "dig deeply" for harmonic distortion--- i.e. QA403 vs. more a traditional analog notch approach? Perhaps QA403 vs. your THD analyzer?
Is your distortion magnifier technique still advantageous when using a QA403 as the signal source? Do super-low THD oscillators (eg. Victor) become unavoidable when exploring state of the art? I recognize the power of spectrum analyzers to dig into the noise, but am focused on limitations arising from purity of the source.
Many thanks for your insights.
Best regards,
Steve
Hi Steve,
Just a few days ago I put the 69-page Version 2 of the QA403 tutorial up on my website at cordellaudio dot com. In it, I show some loop back performance of the QA403 for one particular case where the signal is at 0 dBV. Other signal levels and settings for the full-scale input (here +18 dBV) will yield a bit different performance. One useful measure is the FFT noise floor, above which you can see distortion product lines, including those of the QA403 itself. In the loopback test I showed, the FFT noise floor was an impressive -140 dBV. This soes not mean you can measure your device down that low, due to the distortion of the QA403 itself. Reported THD was about -115 dB or better in this particular loop back test. As an example, in this test, at 1 kHz, 2nd harmonic was -130 dB or better. 3rd harmonic was -120 dB or better. These numbers are competitive with other very good analyzers, and in my opinion more than good enough for anyone but the most serious designers.
My analyzer, being traditional analog without spectral analysis, only does THD+N. At 1 kHz its THD+N residual is about -107 dB. Putting a very good passive or acrive twin T notch in front of it will of course allow much deeper readings. Note that putting a twin T in front of an analyzer like mine requires that the notch depth must have a known limited deepness, such as 40 dB, for the analyzer auto-tune to lock onto and use as a reference level for the fundamental.
Alternatively, one can put a spectrum analyzer or FFT on the residual output of an analyzer like mine. This allows one to see individual THD harmonics pretty far down. I often do this. I have a 1 kHz Victor oscillator and looked at it this way maybe a year ago. I think I was able to see pretty far below -120 dB using an HP3580A spectrum analyzer on the residual.
One advantage of using the Distortion Magnifier (DM) is that it tends to improve measurement in which the oscillator is the dominant analyzer source of distortion, but one must realize there may be some harmonic cancellation in this approach. It is also a bit time consuming to adjust the amplitude and phase of the cancellation. The DM can add 20 or 40 dB of dynamic range to an ordinary analyzer.
With respect to the QA403, I have not looked closely enough to see whether the biggest contributor to its own distortion is the generator or the analyzer.
I recommend the $600 QA403 without reservation. It is a bargain, with very impressive performance.
Cheers,
Bob
Just a few days ago I put the 69-page Version 2 of the QA403 tutorial up on my website at cordellaudio dot com. In it, I show some loop back performance of the QA403 for one particular case where the signal is at 0 dBV. Other signal levels and settings for the full-scale input (here +18 dBV) will yield a bit different performance. One useful measure is the FFT noise floor, above which you can see distortion product lines, including those of the QA403 itself. In the loopback test I showed, the FFT noise floor was an impressive -140 dBV. This soes not mean you can measure your device down that low, due to the distortion of the QA403 itself. Reported THD was about -115 dB or better in this particular loop back test. As an example, in this test, at 1 kHz, 2nd harmonic was -130 dB or better. 3rd harmonic was -120 dB or better. These numbers are competitive with other very good analyzers, and in my opinion more than good enough for anyone but the most serious designers.
My analyzer, being traditional analog without spectral analysis, only does THD+N. At 1 kHz its THD+N residual is about -107 dB. Putting a very good passive or acrive twin T notch in front of it will of course allow much deeper readings. Note that putting a twin T in front of an analyzer like mine requires that the notch depth must have a known limited deepness, such as 40 dB, for the analyzer auto-tune to lock onto and use as a reference level for the fundamental.
Alternatively, one can put a spectrum analyzer or FFT on the residual output of an analyzer like mine. This allows one to see individual THD harmonics pretty far down. I often do this. I have a 1 kHz Victor oscillator and looked at it this way maybe a year ago. I think I was able to see pretty far below -120 dB using an HP3580A spectrum analyzer on the residual.
One advantage of using the Distortion Magnifier (DM) is that it tends to improve measurement in which the oscillator is the dominant analyzer source of distortion, but one must realize there may be some harmonic cancellation in this approach. It is also a bit time consuming to adjust the amplitude and phase of the cancellation. The DM can add 20 or 40 dB of dynamic range to an ordinary analyzer.
With respect to the QA403, I have not looked closely enough to see whether the biggest contributor to its own distortion is the generator or the analyzer.
I recommend the $600 QA403 without reservation. It is a bargain, with very impressive performance.
Cheers,
Bob
It's the D/A (gen side) in the QA403 that is limiting the ultimate performance of the QA403, not the A/D.
Have you enabled cross-correlation (same sig into L+R) for single channel noise measurements yet? (right click the run tab and enable it). Prepare to be amazed.
Of course. Anyway, I always have an itch to set pinpricks to those audiophools 😎.
I didn't mean these elctrolytics and/or DC servos, but the RIAA network itself instead. In my opinion the dielectric properties exclusively are (might be...) of some impact when there's some voltage across the related capacitor, not with a coupling capacitor with negligible impedance across the AF frequency range.
Best regards!
I understand your point, but still shy away from using capacitors of questionable audio quality in the signal path. That feedback shunt electrolytic capacitor is in the signal path, even though it is not part of the RIAA equalization. Actually, you may remember the now-extinct IEC low-frequency corner in the RIAA spec that created a (very) poor man's LF flutter filter. One designer (whos name I will not mention) used that feedback shunt capacitor to implement that rolloff, so that capacitor was REALLY in the signal path. You can also imagine the poor frequency tolerance of that corner frequency unless you used an unobtainium precision electroltic capacitor.
Your point about amount of signal voltage appearing across the electrolytic is a good one. It basically means that if you must use such a capacitor you should use a large one to keep the voltage across it small, even at low frequencies and even if you do not need it to be so large in order to meet your low-frequency roll-off target. I have also found that using NP electrolytic capacitors of significantly higher voltage rating reduces their distortion. Using capacitors designed for use in passive loudspeaker crossovers can help here.
Finally, one other note about using DC servos. In normal practice without a DC servo, it is commonplace engineering to make the DC resistance that both bases of a differential pair see the same, so as to cancel the DC offset effects created by base current, assuming that betas are matched. If one wants to use low impedances to keep the noise down, this can result in amplifier input impedance that is lower than would be desirable, perhaps in the range of 10-30k. This means that the AC coupling capacitor at the input may have to be larger to meet the low-frequency roll-off target. This also can increase the low-frequency loss due to the output coupling capacitor in the preamp (over which you have no control). If instead a DC servo is used to minimize DC offset at the output, one is now free to use non-equal values of DC resistance on either side of the differential pair. This allows one to use a larger base return resistance on the input side of the differential pair without incurring a DC offset penalty.
Cheers,
Bob
Your point about amount of signal voltage appearing across the electrolytic is a good one. It basically means that if you must use such a capacitor you should use a large one to keep the voltage across it small, even at low frequencies and even if you do not need it to be so large in order to meet your low-frequency roll-off target. I have also found that using NP electrolytic capacitors of significantly higher voltage rating reduces their distortion. Using capacitors designed for use in passive loudspeaker crossovers can help here.
Finally, one other note about using DC servos. In normal practice without a DC servo, it is commonplace engineering to make the DC resistance that both bases of a differential pair see the same, so as to cancel the DC offset effects created by base current, assuming that betas are matched. If one wants to use low impedances to keep the noise down, this can result in amplifier input impedance that is lower than would be desirable, perhaps in the range of 10-30k. This means that the AC coupling capacitor at the input may have to be larger to meet the low-frequency roll-off target. This also can increase the low-frequency loss due to the output coupling capacitor in the preamp (over which you have no control). If instead a DC servo is used to minimize DC offset at the output, one is now free to use non-equal values of DC resistance on either side of the differential pair. This allows one to use a larger base return resistance on the input side of the differential pair without incurring a DC offset penalty.
Cheers,
Bob
I also wouldn't use capacitors of questionable quality. My point was that there's next to no significance between any modern film capacitors, as long as there's no siginificant voltage across it. Due to lower dielectric constant, PP capacitors might not fit because they're too big.
OTOH, electrolytics as coupling capacitors are common practice even in renowned studio mixing consoles...
Best regards!
OTOH, electrolytics as coupling capacitors are common practice even in renowned studio mixing consoles...
Best regards!
Thanks Rick for your thoughtful comments. I'll likely short the DC blocker and just use zfoil 100R and 100k to ground for input no ? I want to match input in impedance of Pass Labs for biamp.
Thank you Mr. Cordell. For your reimagined Ernesto Borbely and David Hafler's wonderful circuit. A DC servo is something I've not implemented before..what are the benefits of this if input stage jfets are closely matched ? PS Jim Strickland's transnova is also a thing of beauty to me.. did you know Jim ?
Not long after Erno Borbely’s design of the original Hafler DH 200 and his first DIY amplifiers he moved to DC servos in all his solid state jfet/fet designs for the last 30 years. I use his latest DAC, phono, line amp, and power amp designs, all have DC servos and no coupling caps (with DC protect boards in the power amps) in a fully direct coupled system. The cost and space of even industrial quality Wima coupling caps, let alone Rel Cap, Mundorf, Audio Note etc make a servo both cost and space effective with a lower sonic signature.
Last edited:
Phew. That was a lot of reading, 134 pages. I wanted to read through them before posting. Anyways, I picked up a stock DH-220 with PC-19B boards. I have adjusted the bias and output offset and did some rewiring that included taking the fuse out of the feedback path and installing some hum reducing resistors from point 2 to point 7 on the PCB . The PC-19B does not have C23 between points 2 & 7 like the PC-19C has. I have changed to isolated RCA inputs and use shielded twisted pair from the RCA connection to points 1 & 2 on the PCB. On a side note, the cable from the back of the RCA jack to the PCB is very sensitive to placement. 180 Hz hum with no signal present can be over 10 dB higher if the cable is near the chassis.
My distortion using the QA403 is much higher than the specifications at high frequencies. This is true of the unmodified DH-220 as well. I also have a couple of Sonance amps with much higher distortion than specifications also at high frequencies. Were the original specification done at some magical settings? The original specification at 115W into 8 ohms is 0.0025% at 1 kHz and 0.012% at 20 kHz. I'm seeing distortion levels over 10x these numbers.
On the QA403, I use a 192k sampling rate, 256k FFT, THD frequency from 90 to 80k Hz. I use differential inputs to the QA403 from the speaker binding posts on the amplifier and a short single ended shielded cable from the output from the QA403 to the amplifier. I also connect the amplifier chassis ground to the shield of the USB cable as this makes a huge reduction of high frequency noise and 60 Hz and all it's evil harmonics. I use a non-inductive 8 ohm load.
Here is the distortion of the right channel of my DH-220 with the fuse jumpered at the back of the fuse holder. This was taken right after running the same test with the fuse in the circuit (see below).
Here is the distortion of my DH-220 with a 5A fuse. This was taken after running a couple of other long tests, so it was all warmed up.
As you can see, the distortion with the fuse is much higher than with it jumpered. This mainly occurs at the lower frequencies and higher signal levels. I am going to rewire the output so the fuse is back in the negative feedback loop. +4 dB gives 132 watts of output into my 8 Ohm load.
My question is, is the high frequency distortion typical of amplifiers even it the specifications say it is supposed to be low? i.e. they are either lying or distorting the truth (distorting the distortion).
I read that in these pages that matching the IPS transistors can lower distortion, but that discussion as well as most of the discussions seemed to be for the DH-200. IDK if the IPS transistors were matched from the factory on the DH-220.
I tested the power supply capacitors and one of them tested at 9100uF with an ESR of 0.054 ohms and the other one tested at 9500 uF with an ESR of 0.052 Ohms. This gives a D.F. of about 0.37 for these old capacitors.
My distortion using the QA403 is much higher than the specifications at high frequencies. This is true of the unmodified DH-220 as well. I also have a couple of Sonance amps with much higher distortion than specifications also at high frequencies. Were the original specification done at some magical settings? The original specification at 115W into 8 ohms is 0.0025% at 1 kHz and 0.012% at 20 kHz. I'm seeing distortion levels over 10x these numbers.
On the QA403, I use a 192k sampling rate, 256k FFT, THD frequency from 90 to 80k Hz. I use differential inputs to the QA403 from the speaker binding posts on the amplifier and a short single ended shielded cable from the output from the QA403 to the amplifier. I also connect the amplifier chassis ground to the shield of the USB cable as this makes a huge reduction of high frequency noise and 60 Hz and all it's evil harmonics. I use a non-inductive 8 ohm load.
Here is the distortion of the right channel of my DH-220 with the fuse jumpered at the back of the fuse holder. This was taken right after running the same test with the fuse in the circuit (see below).
Here is the distortion of my DH-220 with a 5A fuse. This was taken after running a couple of other long tests, so it was all warmed up.
As you can see, the distortion with the fuse is much higher than with it jumpered. This mainly occurs at the lower frequencies and higher signal levels. I am going to rewire the output so the fuse is back in the negative feedback loop. +4 dB gives 132 watts of output into my 8 Ohm load.
My question is, is the high frequency distortion typical of amplifiers even it the specifications say it is supposed to be low? i.e. they are either lying or distorting the truth (distorting the distortion).
I read that in these pages that matching the IPS transistors can lower distortion, but that discussion as well as most of the discussions seemed to be for the DH-200. IDK if the IPS transistors were matched from the factory on the DH-220.
I tested the power supply capacitors and one of them tested at 9100uF with an ESR of 0.054 ohms and the other one tested at 9500 uF with an ESR of 0.052 Ohms. This gives a D.F. of about 0.37 for these old capacitors.
Distortion is always higher at high frequencies. There are good reasons for this, so no worries.
👍
Guess what, people who don’t realize the DC resistance at differential input needs to be matched, they can’t do DC servo right, either. Although opamps have lower input bias current, the DC resistance should also be matched to get DC offset down.
PS: I am talking about non-FET type amps or opamps.
Putting the speaker fuse inside the global feedback loop is usually a good idea. The fuse wire inside the fuse is usually designed to exhibit a fairly high temperature coefficient to enable the fuse to blow as desired. This leads to the distortion it can cause at low frequencies, since its resistance is changing with signal current when enough time elapses over a cycle of the signal. For some amplifiers, it can be a concern that when the fuse opens and is in the feedback loop, the amplifier goes open loop with very large open-loop signal excursions, even with a small signal or no signal at all. This situation can be undesirable for some amplifiers, and might conceivably cause some damage. It is possible to design circuitry that essentially keeps the loop closed to some extent when the fuse opens. This is addressed in my power amplifier book. BTW, the same type of issues can be in play when there is an output mute or protection relay and one includes it in the global feedback loop to reduce distortion from the relay.
Version 3 of my Tutorial for the QuantAsylum QA403 audio analyzer is now up on my website and is discussed in the QuantAsylum User Forum. The tutorial walks the reader through all of the pertinent tests for an audio power amplifier, in this case my BC-1 design, which is also described in detail on my website.
The QA403 is an outstanding piece of equipment, especially at only $600. It has pretty much all of the functionality and performance most people could need. It is just as valuable for use with small-signal circuits as well as power amplifiers. One nice feature is that it includes selectable inverse RIAA respnse shaping for measuring phono preamps. It also includes selectable A-weighting for noise measurements. Further, it includes optional user-entered weighting responses. It can usually see distortion products below -120 dB or better, and report THD down to -120 dB or better. It can sample at up to 192 kHz, allowing one to see FFT products up to at least 80 kHz. Sampling to 384 kHz may be available soon. It is powered from its USB connection to the PC on which its software resides, and it is galvanically isolated from the PC side. It has balanced I/O.
Cheers,
Bob
If you put the fuse inside the feedback loop, you should shunt some signal across the fuse so the amplifier has at least some feedback.
In general, a fuse isn't really speaker protection. It adds nonlinear impedance and doesn't do much to save speakers. There is a time constant at work here. If you want to protect speakers, do it effectively. You can use a relay (the best) or shut down the power supply (Carver did this very well). Before people start crying, a relay is a maintenance device. If you abuse them, you change them more often. Relays in my equipment last decades. Of course, you can do what one manufacturer of a very high power amplifier did. Crowbar protection. It was effective and saved the speaker stack. It also caused very impressive damage to wire and PCB foil. Semi's lost their legs in fact.
In general, a fuse isn't really speaker protection. It adds nonlinear impedance and doesn't do much to save speakers. There is a time constant at work here. If you want to protect speakers, do it effectively. You can use a relay (the best) or shut down the power supply (Carver did this very well). Before people start crying, a relay is a maintenance device. If you abuse them, you change them more often. Relays in my equipment last decades. Of course, you can do what one manufacturer of a very high power amplifier did. Crowbar protection. It was effective and saved the speaker stack. It also caused very impressive damage to wire and PCB foil. Semi's lost their legs in fact.
You are exectly right on all counts. Nevertheless, speaker fuses are used in some amps. There are some different ways of keeping the FB loop closed somewhat when the fuse is blown. Using a crowbar is excellent as long as the amplifier has good short-circuit protection for all conditions. A crowbar alone does not provide an option for muting like a relay does.
In some amplifier designs I think you can turn off the output stage electronically by collapsing the bias spreader with an SCR driven by an opto-isolator, but this does not protect the loudspeaker if the output transistors have shorted. Some have looked at solid state relays, but some of those arrangements may introduce distortion. I've never tried it, but trench MOSFETs with very low Rds_on might work well if controlled properly.
Cheers,
Bob
In some amplifier designs I think you can turn off the output stage electronically by collapsing the bias spreader with an SCR driven by an opto-isolator, but this does not protect the loudspeaker if the output transistors have shorted. Some have looked at solid state relays, but some of those arrangements may introduce distortion. I've never tried it, but trench MOSFETs with very low Rds_on might work well if controlled properly.
Cheers,
Bob
Hi Bob,
Yes, fuses are popular because they are cheap. Period. They are attractive after the bashing relays have had (undeserved) from some irresponsible companies. Some of the first amplifiers I serviced in the 1970's were "fuse protected" on the speaker output. I modified some to include the fuse in the feedback loop, but I split the feedback points to look after shop when the fuses opened. We still have some manufacturers doing this 50 years later.
Killing the bias only helps to minimize turn on pops, not that great at that either. The only ways to protect the load is to either fully disconnect the output, or short it. Shorting is brutal, but extremely effective. It typically causes output stage damage, or further output stage damage. Crowbar protection releases as soon as you remove the triac drive, same as a contactor (heavy relay) would. However, it is common the triac has turned into a piece of wire.
I think the only non-violent but very effective form of speaker protection that isn't a relay was perfected by Carver. You can have a shorted output (impressive on a 375 wpc / 8R amp) and not damage a speaker or cause further damage in the amplifier. It shuts the power supply down quickly. The problem here is that the speaker often expired earlier due to excessive dissipation.
Yes, fuses are popular because they are cheap. Period. They are attractive after the bashing relays have had (undeserved) from some irresponsible companies. Some of the first amplifiers I serviced in the 1970's were "fuse protected" on the speaker output. I modified some to include the fuse in the feedback loop, but I split the feedback points to look after shop when the fuses opened. We still have some manufacturers doing this 50 years later.
Killing the bias only helps to minimize turn on pops, not that great at that either. The only ways to protect the load is to either fully disconnect the output, or short it. Shorting is brutal, but extremely effective. It typically causes output stage damage, or further output stage damage. Crowbar protection releases as soon as you remove the triac drive, same as a contactor (heavy relay) would. However, it is common the triac has turned into a piece of wire.
I think the only non-violent but very effective form of speaker protection that isn't a relay was perfected by Carver. You can have a shorted output (impressive on a 375 wpc / 8R amp) and not damage a speaker or cause further damage in the amplifier. It shuts the power supply down quickly. The problem here is that the speaker often expired earlier due to excessive dissipation.
If I don't have the speaker protection, I just put low wattage 10 ohm resistor on the transformer center tap before the filter caps. As AC is shunted by large filter caps, there is only small DC current flowing on the transformer center tap. The resistor restricts any large current over the center tap. The principal is simple. Amps would not be able to source any DC if the transformer center tap is open (not connected).
Last edited:
Depends on the supply type. A bipolar supply for solid state can have a massive amount of load current flowing in the centre tap. It doesn't cancel. The current waveform is peak current spikes on idle.
In tube amps you describe the "standby switch". I wouldn't want to place a resistor in there either.
In tube amps you describe the "standby switch". I wouldn't want to place a resistor in there either.