Interesting read on AES/Engineering observations of tubes and transistors

[Geardaddy]

Totally agree, Geardaddy...i also would love to read an update of this article. BTW, i did not know about the Goldmund high bandwidth thing...i think Gryphon does that as well...flat to 350khz or something.

I think Goldman amplifiers extend out to 3 MHz

 

[Geardaddy]

Another article I love on the emotive side of this subject that should be replicated today....

http://www.stereophile.com/features/203/

 

[LL21]

I think Goldman amplifiers extend out to 3 MHz

yikes!

 

[esldude]

Well early Spectral gear was good to a megahertz. Have tested some and it is no lie. Later Spectral gear is good to 2 mhz (unless they have moved that even further in recent times).

In the Spectral case, some of the designers have indicated they wanted an amp that releases the signal cleanly and quickly after transients. Some SS power transistors on transient music signals will have the exact bias point disturbed by heating effects of the junctions of the transistors enough to alter the circuit reaction. And this usually would also cause the amp not to let go of signal peak as quickly as the input signal might. I don't know the full details, but those guys have said the wide, wide bandwidth was not a target whatsoever. But getting an amp to release the signal quickly and not have momentary shift of bias at the transistor junction caused the design to be wide bandwidth as a side effect. Whether that is the case with Goldmund I don't know.

If not for this effect an SS amp -3 db at 200 khz should be plenty flat to 20 khz. Spectrals certainly sound like they let go of and keep up with a signal to me. Some of the better Class D amps seem to do this as well (when judged subjectively) without having any extra wide bandwidth.

 

[Tony Lauck]

Do not assume that acoustic music can be played back without clipping. This will depend on the speakers and the room as well as the type of acoustic music. Mahler Symphonies and pipe organ music can be very power intensive if one attempts playback at realistic row 10 concert volumes. Instantaneous peak SPLs can be around 120 dB even though the average level may be 30 dB lower for FFF peaks.

As to output voltages on digital playback. For example, assume a gain setting where a 1 kHz sinewave that peaks at 0 dBfs produces an output signal that peaks at + - 1.0 volts. It is possible for the analog output on music recordings to exceed this level by several decibels as the result of "inter sample peaks". The analog playback circuits (and digital upsampling circuits if applicable) needs a certain amount of headroom to avoid clipping. It is possible to create bizarre test examples of digital files where peaks can be over 10 dB greater than 0 dBfs. The peak voltage would be over 3 volts for this test signal in the example case, assuming that there was sufficient headroom in the DAC circuitry. (I have one such file, but I won't share it, because although it will be inaudible it will do an excellent job of smoking some tweeters.)

 

[microstrip]

Do not assume that acoustic music can be played back without clipping. This will depend on the speakers and the room as well as the type of acoustic music. Mahler Symphonies and pipe organ music can be very power intensive if one attempts playback at realistic row 10 concert volumes. Instantaneous peak SPLs can be around 120 dB even though the average level may be 30 dB lower for FFF peaks.

As to output voltages on digital playback. For example, assume a gain setting where a 1 kHz sinewave that peaks at 0 dBfs produces an output signal that peaks at + - 1.0 volts. It is possible for the analog output on music recordings to exceed this level by several decibels as the result of "inter sample peaks". The analog playback circuits (and digital upsampling circuits if applicable) needs a certain amount of headroom to avoid clipping. It is possible to create bizarre test examples of digital files where peaks can be over 10 dB greater than 0 dBfs. The peak voltage would be over 3 volts for this test signal in the example case, assuming that there was sufficient headroom in the DAC circuitry. (I have one such file, but I won't share it, because although it will be inaudible it will do an excellent job of smoking some tweeters.)

Tony,

I have read about this "inter sample peaks" some time ago in another forum, but I have never seen any at the output of a typical CD player or DAC playing acoustic CD recordings. At that time I looked for them it using a Tektronix digitizing scope with peak detecting function sampling a 1MHz and never was able to detect any of them. Do you have any example of any existing recording of acoustic music having such peaks? Perhaps this effect is more significant at the recording and mastering phase in the analog and digital domain.

Is there any DAC analog output filter topology that is more susceptible to display those peaks?

 

[Tony Lauck]

Tony,

I have read about this "inter sample peaks" some time ago in another forum, but I have never seen any at the output of a typical CD player or DAC playing acoustic CD recordings. At that time I looked for them it using a Tektronix digitizing scope with peak detecting function sampling a 1MHz and never was able to detect any of them. Do you have any example of any existing recording of acoustic music having such peaks? Perhaps this effect is more significant at the recording and mastering phase in the analog and digital domain.

Is there any DAC analog output filter topology that is more susceptible to display those peaks?

The peaks are caused by the filtering action and the peaks can happen with either digital or analog filtering. The issue can occur during recording or playback, however presumably problems during recording can be eliminated by adjusting the levels during the production process. A simple example will show how filtering can increase the peaks. Imagine a square wave at a given peak level. If you filter out the harmonics the result will be a sine wave at the same frequency. However, the sine wave will increase in amplitude by a factor or 4 / pi, 2.1 dB.

Properly made recordings should not have lots of energy at the filter cut-off frequency used in playback. However, there being no standards in this regard, anything goes. Pro equipment and some audiophile playback software allows for digital gain control and this will eliminate the problem in playback. I use HQPlayer and it will display a counter of how many samples got clipped due to inter sample peaks. Usually one or two samples will not be audible, but if the counter starts counting the music sounds harsh unless I fix things by reducing the digital gain.

 

[bonzo75]

Well early Spectral gear was good to a megahertz. Have tested some and it is no lie. Later Spectral gear is good to 2 mhz (unless they have moved that even further in recent times).

In the Spectral case, some of the designers have indicated they wanted an amp that releases the signal cleanly and quickly after transients. Some SS power transistors on transient music signals will have the exact bias point disturbed by heating effects of the junctions of the transistors enough to alter the circuit reaction. And this usually would also cause the amp not to let go of signal peak as quickly as the input signal might. I don't know the full details, but those guys have said the wide, wide bandwidth was not a target whatsoever. But getting an amp to release the signal quickly and not have momentary shift of bias at the transistor junction caused the design to be wide bandwidth as a side effect. Whether that is the case with Goldmund I don't know.

If not for this effect an SS amp -3 db at 200 khz should be plenty flat to 20 khz. Spectrals certainly sound like they let go of and keep up with a signal to me. Some of the better Class D amps seem to do this as well (when judged subjectively) without having any extra wide bandwidth.

Hi those who have heard the Spectral, have you heard the Sanders Magtech? How does it compare

 

[marty]

I think Goldman amplifiers extend out to 3 MHz

Its not just a matter of bandwidth that seems to be important in determining an amplifier's sound. One issue that seems to be equally important is the role of global negative feedback. In 1973, a young Finish engineer, Dr. Matti Otala presented to the Audio Engineering Society a paper in which he described a distortion mechanism called transient intermodulation, or TIM, and which had otherwise gone undetected, apparently because the standard measurements of the day relied using mainly steady-state test tones. Dr. Otala proved that the nonlinearities he described were audibly present in solid-state amplifiers that used global feedback to reduce other, more commonly known distortions. Aside from the discovery of TIM, Otala is credited with exposing the negative effects of negative global feedback in amplifier design. Otala was the head engineer Harmon Kardon from 1978-80 (he now consults for Electrocompaniet and was for many years, the chief engineer at Nokia), and designed the Citation Amplifier, thought by many to be a fine sounding amplifier in its day. His main tenant was that the sound of an amplifier was best correlated with the bandwidth of an amplifier with open loop gain; that is, no negative feedback, which he eschewed. As far as I know, this was a principle that was rapidly accepted by designers of solid state amplifiers for several decades. However, many of the Swiss companies who are now members of the ultra megahertz bandwidth school (i.e. Goldmund, Solution, CH) seem to have turned this concept on its head and flaunt designs in which it is argued that negative bandwidth is not necessarily a bad thing but in fact a good thing. For example, the Solution amps (Which Valin an Harley both seem to love) is characterized by high negative feedback circuitry. I'd love to know what Otala thinks of these current designs, and whether there is something about their design that makes them more or less immune to the conventional wisdom for many years that suggested high negative feedback designs were something best to avoid.

 

[esldude]

Hi those who have heard the Spectral, have you heard the Sanders Magtech? How does it compare

I have not heard the Sanders Magtech. I use electrostats and was interested in his ESL amps at one time. I heard them briefly once with some Martin Logans. Not enough to form an opinion other than the amps seemed fine not drawing attention to themselves in a cursory listen.

Was not familiar with the Magtech so looked it up. Sounds like a variation on voltage rail switching that Carver used at one time. Probably more sophisticated and updated, but a similar concept. As to how it sounds I could not say. If you are contemplating them, you could always do the 30 day home trial.

 

[esldude]

Its not just a matter of bandwidth that seems to be important in determining an amplifier's sound. One issue that seems to be equally important is the role of global negative feedback. In 1973, a young Finish engineer, Dr. Matti Otala presented to the Audio Engineering Society a paper in which he described a distortion mechanism called transient intermodulation, or TIM, and which had otherwise gone undetected, apparently because the standard measurements of the day relied using mainly steady-state test tones. Dr. Otala proved that the nonlinearities he described were audibly present in solid-state amplifiers that used global feedback to reduce other, more commonly known distortions. Aside from the discovery of TIM, Otala is credited with exposing the negative effects of negative global feedback in amplifier design. Otala was the head engineer Harmon Kardon from 1978-80 (he now consults for Electrocompaniet and was for many years, the chief engineer at Nokia), and designed the Citation Amplifier, thought by many to be a fine sounding amplifier in its day. His main tenant was that the sound of an amplifier was best correlated with the bandwidth of an amplifier with open loop gain; that is, no negative feedback, which he eschewed. As far as I know, this was a principle that was rapidly accepted by designers of solid state amplifiers for several decades. However, many of the Swiss companies who are now members of the ultra megahertz bandwidth school (i.e. Goldmund, Solution, CH) seem to have turned this concept on its head and flaunt designs in which it is argued that negative bandwidth is not necessarily a bad thing but in fact a good thing. For example, the Solution amps (Which Valin an Harley both seem to love) is characterized by high negative feedback circuitry. I'd love to know what Otala thinks of these current designs, and whether there is something about their design that makes them more or less immune to the conventional wisdom for many years that suggested high negative feedback designs were something best to avoid.

I think it is Bruno Putzeys who said the negative effects of negative feedback are from misuse and not using enough. You can read his explanation here:


http://www.edn.com/design/consumer/...fiers--Why-there-is-no-such-thing-as-too-much

http://www.edn.com/design/consumer/...y-there-is-no-such-thing-as-too-much--Part-2-

Part 2 may be all you really need to read.

 

[Atmasphere]

One question I have for engineer types like Ralph relates to "timing errors" at the amplifier stage. I know that Goldmund, a Swiss engineering based company, makes the point that "high bandwidth" amplifiers are somehow more resistant to time and phase errors intrinsic to the amplifier itself. True or false? I would be interesting to test this with the vertex software to see if such an amplifier translated into lower "error" rates....

True. Bandwidth is required in order to reproduce phase correctly.

Well early Spectral gear was good to a megahertz. Have tested some and it is no lie. Later Spectral gear is good to 2 mhz (unless they have moved that even further in recent times).

In the Spectral case, some of the designers have indicated they wanted an amp that releases the signal cleanly and quickly after transients. Some SS power transistors on transient music signals will have the exact bias point disturbed by heating effects of the junctions of the transistors enough to alter the circuit reaction. And this usually would also cause the amp not to let go of signal peak as quickly as the input signal might. I don't know the full details, but those guys have said the wide, wide bandwidth was not a target whatsoever. But getting an amp to release the signal quickly and not have momentary shift of bias at the transistor junction caused the design to be wide bandwidth as a side effect. Whether that is the case with Goldmund I don't know.

If not for this effect an SS amp -3 db at 200 khz should be plenty flat to 20 khz. Spectrals certainly sound like they let go of and keep up with a signal to me. Some of the better Class D amps seem to do this as well (when judged subjectively) without having any extra wide bandwidth.

Transistors exhibit a phenomena of bias drift due to heat, which can show up in a junction on an instantaneous level. One way around the problem is extremely wide bandwidth.

interesting. bias stability of output devices aside, IMO flat to 20khz at any power form milliwatts to full power is enough for most forms of music as comes to us via recorded mediums. In fact, less bandpass and FR tailoring gets more emotion into the music IMO. I don't believe that amplifying a small part of the lower RF spectrum is a good idea, best to let it die right at the input to the amp.

Some of this is correct, especially the RF issues. However to get phase correct the engineering rule of thumb is 1/10th the lowest frequency to be played on the bottom (2Hz if 20Hz is desired) and 10x the highest frequency (200KHz if 20KHz is desired). Preventing phase shift is important to reproducing the stage width and depth as our ears use phase to detect location. This bandwidth is important in electronics even if there is no signal to be reproduced outside the audio passband.

I am not against feedback but I am against its improper use. Quite often feedback is responsible for injecting RF into amps and preamps- in so doing the effects are deleterious. Feedback will also cause an amplifier to have harsher clipping characteristics even if done correctly. The traditional feedback formulae taught in school are insufficient especially in amplifier design- to do it right at the least one needs a lot of simulation in PSpice or some pretty good math chops, as there are more variables than taught.

 

[DonH50]

Not all wideband circuits are immune to bias drift, and I am not sure broadband feedback corrects bias drift. I must have missed something. Thermal time constants vary widely from the junction through the case (generally ~microseconds to seconds). Thermal stability is a design goal, natch, but IME involves more than just biasing.

Middlebrook (I think) did a nice analysis on the "complete" feedback circuit and analysis and the method used to be included in the app book that came with PSpice. Not sure, I rarely used PSpice (used many other simulators that do the same thing, mostly integrated with our layout tools from Cadence, Mentor, etc.), and I got the procedure from a technical paper. A cheap free simulator for those wishing to play around is LTSpice from Linear Technology. I am currently using it for simple stuff and examples for students and field engineers (HSpice and ADS for real work).

Agree 1000% on the improper design of feedback loops. It is quite possible to make things worse by adding feedback. Some of my best oscillators are amplifiers with improper feedback...

 

[Atmasphere]

Not all wideband circuits are immune to bias drift, and I am not sure broadband feedback corrects bias drift. I must have missed something. Thermal time constants vary widely from the junction through the case (generally ~microseconds to seconds). Thermal stability is a design goal, natch, but IME involves more than just biasing.
.

I agree 100%- its not a great solution, but it helps if you have the bandwidth. As you point out, compensation from outside the device takes time! I'm forgetting now who came up with a solution for this that actually works and does not need the bandwidth.

 

[microstrip]

(...) Transistors exhibit a phenomena of bias drift due to heat, which can show up in a junction on an instantaneous level. One way around the problem is extremely wide bandwidth.(...)

Signal-induced thermal drifts in transistors (also called Memory distortion) were studied by Gérard Perrot in the 70's. We can find information about his work at http://peufeu.free.fr/audio/memory/. The original articles were published in the magazine Audiophile (in french) .

 

[Atmasphere]

^^ Thanks! That's it!

 

[LL21]

A quotation from Thrax the manufacturer in a Mono and Stereo article:

"Why do we use triodes?
Audiophiles have been led to believe through published measurements at maximum output power and uneducated reviewers that single-ended triode amplifiers produce vast amounts of harmonic distortion. As a matter of fact triode vacuum tubes are by far the most linear amplifying devices in existence today. They produce the least amount of distortion, and that distortion is predominately second harmonic, which is the least obtrusive type for the sound. By contrast, pentodes produce greater distortion, and the third harmonic tends to dominate. A transistor looks at best like a very bad pentode."

(http://www.monoandstereo.com/2013/06/thrax-audio-dionysos-preamplifier-and_24.html)

I am personally NOT seeking in a debate out SQ of tubes vs ss...simply a technical explanation (for a non-techie like me) about whether the above statement has reasonable merit. I have read elsewhere posts which suggest part of this may be true, particularly about distortion levels of SETs within their ideal power range (ie, with super-efficient speakers). I expect SOME may be marketing...but nevertheless, i do find it interesting to learn more if this has reasonable merit. Thanks in advance...

 

[Orb]

Not all wideband circuits are immune to bias drift, and I am not sure broadband feedback corrects bias drift. I must have missed something. Thermal time constants vary widely from the junction through the case (generally ~microseconds to seconds). Thermal stability is a design goal, natch, but IME involves more than just biasing.

Middlebrook (I think) did a nice analysis on the "complete" feedback circuit and analysis and the method used to be included in the app book that came with PSpice. Not sure, I rarely used PSpice (used many other simulators that do the same thing, mostly integrated with our layout tools from Cadence, Mentor, etc.), and I got the procedure from a technical paper. A cheap free simulator for those wishing to play around is LTSpice from Linear Technology. I am currently using it for simple stuff and examples for students and field engineers (HSpice and ADS for real work).

Agree 1000% on the improper design of feedback loops. It is quite possible to make things worse by adding feedback. Some of my best oscillators are amplifiers with improper feedback...

Will be interesting to see how many adopt the THX feed forward amp AB/H topology; Achromatic Audio Amplifier.
So far Benchmark has released a product albeit with switchmode power supply rather than transformer (which I think is what is specified by THX).
Anyway the design is pretty elegant and seriously efficient in terms of thermal-low bias requirements.
Also I think it has the best measured S/N that HifiNews has ever measured, A-wtd is over 105dB at 0dBW.

Be interesting to see what others can do with the technology.
Cheers
Orb

 

[DonH50]

A quotation from Thrax the manufacturer in a Mono and Stereo article:

"Why do we use triodes?
Audiophiles have been led to believe through published measurements at maximum output power and uneducated reviewers that single-ended triode amplifiers produce vast amounts of harmonic distortion. As a matter of fact triode vacuum tubes are by far the most linear amplifying devices in existence today. They produce the least amount of distortion, and that distortion is predominately second harmonic, which is the least obtrusive type for the sound. By contrast, pentodes produce greater distortion, and the third harmonic tends to dominate. A transistor looks at best like a very bad pentode."

(http://www.monoandstereo.com/2013/06/thrax-audio-dionysos-preamplifier-and_24.html)

I am personally NOT seeking in a debate out SQ of tubes vs ss...simply a technical explanation (for a non-techie like me) about whether the above statement has reasonable merit. I have read elsewhere posts which suggest part of this may be true, particularly about distortion levels of SETs within their ideal power range (ie, with super-efficient speakers). I expect SOME may be marketing...but nevertheless, i do find it interesting to learn more if this has reasonable merit. Thanks in advance...

A little truth and a lot of emphasis from marketing. The writing is clearly slanted to make you think everything but a triode is "bad".

Differential designs (most SS; it is easy to double the transistor count in a chip, less so with a tube circuit) ideally cancel even-order harmonics, leaving odd-order harmonics dominate, which usually means the third harmonic. Single-ended circuits (like SETs) using any device will generate the entire distortion series and thus second harmonic distortion tends to dominate. That has nothing to do with tubes or transistors; any imperfect (nonlinear device) will exhibit those characteristics.

As for triodes being "the most linear devices in existence today", well, that sounds like hyperbole to me. Compared to what device, at what bias and power level, etc.? What is true is that the distortion series of an ideal tube is factorial while a bipolar transistor's series is exponential. That means that fundamentally a tube can have lower distortion than a bipolar transistor. An ideal MOSFET only has the second-order term and it stops there, so in theory a MOSFET actually has the lowest distortion of any device. In the real world it is moot as design and use will determine the actual distortion.

Any device biased properly and used within its linear range will produce inaudible distortion. IMO the problem with triode power amplifier circuits is their low power output and low efficiency. Add low or no global feedback, seemingly a popular marketing bullet these days, and you get high output impedance and higher distortion driving a speaker to reasonable volume levels given most speakers. The ones who can benefit are those with highly-efficient speakers who sit close to them and/or listen at modest volumes. (I am sure to be blasted by all the exceptions but I think that statement is true for the majority.)

A SET (or any tube amp) will produce "vast amounts of harmonic distortion" when overdriven. So will a transistor amp (bipolar, MOSFET, or any other device). And any of them can sound great in your system depending upon your components and listening tastes. IMO/IME the key is finding a combination of components that sound best to you. Nothing else really matters.

Snowplow just drove by, time to get the snowthrower out and clear the foot or so on the drive and shovel the few feet of snowbank left by the plow in front of the mailbox...

 

[LL21]

Thank you, Don! Yes, that makes sense to me...use the amplification technology well within the confines of its own specification and enjoy low distortion from that amp...and then presumably move on to focus on reducing far greater sources of distortion (room, speakers, vibration, etc).