Welcome to Solid State Guitar Amp Forum | DIY Guitar Amplifiers. Please login or sign up.

March 28, 2024, 07:46:44 AM

Login with username, password and session length

Recent Posts

 

Power amps - Simple or complex?

Started by teemuk, April 14, 2006, 11:17:16 AM

Previous topic - Next topic

teemuk

There are lot of claims that simpler power amplifier circuits sound better than complex modern ones. Could this really be so?

We all know that simplism has some non-arguable benefits:
- The devices are more likely closely matched since there is a smaller amount of them.
- The layout is simpler which can have some benefits: ie. shorter tracks, less interference. (Assuming everything is done right of course).

...As well as drawbacks:
- Simple design is more likely much more unstabile.
- Simple design is more likely much more unlinear.
- Bad layout will more likely cause more problems.



I simulated two circuits in ltspice. (See the thumbnail link). They are almost "identical", meaning the signal path is almost identical. Both circuits use long tailed pair input with the same amount of feedback so the gain is same, at least on 1 kHz. (There's soon more about this). The output topologies are a bit different: The simple, "old-fashioned" circuit has a quasi-complementary output, which was very common until high power complementary transistors became available. The topology has an advantage of closer matching between the output devices, while the "modern" one uses a fully complementary output. Besides output topology and replacing constant resistances or bootstraps with constant current sources these two circuits are identical.

Their performance however is not:
- The "modern" circuit has a more constant gain throughout bandwidth. The gain of the simple circuit drops heavily on higher frequencies.
- "Modern" circuit can swing closer to rails.
- Simple one suffers from higher DC offset.
- Simple one has worse distortion figures, especially on 2nd harmonic. An exception occures when the circuits are hardly overdriven: In the simpler circuit the 2nd harmonic drops 10dB in relation to "modern".

Every aspect of the simulation shows that the modern circuit is far more better. It shouldn't be more sterile or colder sounding, nor does it clip harsher in relation to simpler one. As a matter of fact, the the simpler one clips harsher than the modern! Undoubtedly the real-life indifferencies between devices will add something to the game. Is it enough to make a simple circuit perform better - i don't think so. Could this preferance over simplism be explained by "psychoacoustics": One knows that he is listening to a simpler circuit and regards the distortion as "pleasentness"? I will soon do some listening tests between these two simulated circuits and get back to this.

Meanwhile, be free to comment and present your ideas.

Stompin_Tom

Wow. Cool idea Teemuk. I've been wondering about the relative performance of various power amp designs... So, the basic difference between the 'simple' and 'modern' circuits is four transistors? Can you explain simply what they do in the circuit?

teemuk

#2
Well, i'm not very good at explaining everything in a simple form but i can try.

The job of Q18 and Q11 (plus the circuitry that's surrounding them) is to provide a constant current for transistors. Why constant current? Well, If we take a basic common emitter circuit the voltage at it's transistor's collector can be calculated from the formula

      Vc = Vcc – IcRc,

which is simply the subtraction of supply voltage and voltage drop over collector resistor. Despite of it's simplicity the formula shows two very important things: 1. Fluctuations in the supply voltage show up in the collector voltage and 2. Smaller voltage drop over a constant resistance means a smaller collector current Ic. The latter means that as the transistor's collector voltage increases the collector current must decrease since the voltage difference between the collector and the supply becomes smaller. If the collector current decreases it means that the emitter current has to decrease as well and less emitter current means lower gain. The exact opposite happens when the output voltage decreases and as a result the waveform will become distorted having a flattened top halfwave and a stretched bottom halfwave.

So, the idea is simply to provide a constant current for the transistors to retain their linearity. If the current over resistor is

      I = U / R,

one can see it will fluctuate within the value of U. Constant current, however, remains constant. The bootstrap (R12, R1 & C4) is a form of constant current source too. It tries to create a constant (DC) voltage drop (= constant current) across the "lower" resistor (R1). (Ideally the AC voltage at the output should equal AC voltage at Q1's collector). The constant current actually allows the "higher" end of the capacitor to swing past rail voltage, which means the voltage at the output will swing closer to supply rail. Modern circuits tend to replace this form of CCS with an active one. Why? It is more linear and doesn't require a bulky capacitor.

Transistors Q17 and Q16 are another form of constant current source. Their job is to balance the loading of the differential circuit, which results into more linear operation.

I hope this was simple enough. If you have more interest, Rod Elliott has written some good articles about the operation of constant current sources and basics of amplifier design:

http://sound.westhost.com/ism.htm
http://sound.westhost.com/amp_design.htm

skey

I would be very interested to see the results of your listening tests, but what will the subjects be listening to?

I think it would vary if they listened to an electric guitar straight in, verses music.
For music, at reasonable levels I highly doubt there will be much of a difference.
For guitar, a clean transparent amp would probably give the "sterile" sound as really that what comes out of a e-guitar.   The cheap amp my sound better as it will give some colorization to the guitar.
On the other hand, if you had a processed guitar signal going in, then I'd say you want a transparent amp so you don't add any extra colorization than what you put in.

All just guessing, it'll be interesting to see the results of the tests.

teemuk

#4
Unfortunately i currently have no have time to construct these circuits in real-life so i have to rely on spice simulation that doesn't guarantee accurateness. However, ltspice has a feature that allows feeding the circuit with a .wav file and extracting the output. Unfortunately, this means that i will loose the "touch" to the guitar. In some sense this is a good thing: I do not have a chance to adjust my playing to the circuit. Basically, i will try this and also see if i can create some variation between identical spice components in order to simulate the real-life differencies that they might have. I have to find out how to do this correctly without biasing the results.

As of the test signals... well, naturally i have been planning to try with both "clean" and processed. Clean signal should be a good measure since it exhibits some transients that should overdrive the power stage and test how much the clipping changes the overrall tone. Currently unclipped signals of both circuits seem identical, except for the different frequency response. Processed tone... Well, i was planning to use some kind of distorted tone since it contains sharp edges that should - or should not - be amplified accurately. Also, both signals (clean and distorted) retain a lot of variations that should be tested: Ie. Palm muting, pinch harmonics, using a pickup or just finger picking. Trying both signal types with reverb is also one of my plans since producing reverb seems to be difficult for some systems.

And no. I have not planned to test the circuits with ie. different instrument signals or with a music or speech source. I already know from experience that the purpose of the amplifier dictates a lot of how it should behave. I feel that including any other than guitar amplification purpose to the test would expand the spectrum too much and i plan to keep this project as a way of measuring the production of guitar sound.

Colorization is a good point. Partly because i run a simulation both circuits are quite "flat", meaning the output is free from the colorisation, created by, for example the guitar speaker, that could actually change the tone drastically. (Excluding the simulation of speaker's impedance of course). I have a way to simulate the guitar speaker's frequency response so i can include it to the test. Anyway, i feel this should be done so that it will not bias the results.

I still have to plan how i will compare the results since i probably can not gather a test group. I have to find a way to play the clips in a random order (that i do not know at first). Then during listening i will write down my impressions and later see if they had any commonities to the the circuit's that produced the soundclips. The drawback of this test is that it's results are based on to my ears. They will be dictated by my taste, no matter what i do. If i find a way to upload some sound samples to the internet i will do it.

Edit: I will probably also create some clips where the output signals are mixed to some musical context.

teemuk

More simulation results and some impressions of my first - and non-blinded - listening tests:

- First and only easy-to-detect difference between the two circuits is that the simpler one exhibits a noticeable startup thump while voltages settle.
- Overdriving creates some differences between output signals. (See the first post). However, these differences seem to be detectable only by visual comparison of both output signals and at least i couldn't hear any of them.
- Just as i guessed, temperature variations affect the simple design more: Distortion varies within few decibels and gain varies few millivolts. (Max. 0.005% variation in the output). Modern design seems to be much more stabile. LTspice can not simulate a thermal breakdown, which would probably occure faster than any noticeable difference between output signals could be detected. Interesting point is that when the ambient temperature rises very high (100 C) the frequency response of both amplifiers becomes almost identical. Altogether, the differences caused by temperature changes where a lot smaller than i thought they would have been.
- Subtle differences between same transistor models seem to cause no measurable difference to the performance. I could push this test even further but i feel that would be deliberate biasing of the results. Usually the transistor's should be well matched anyway.

Actually, i'm very amazed that it's really hard to detect any difference between the circuits. I could be missing some important point. Please do comment and recommend some ideas.

R.G.

Albert Einstein said that everything should be as simple as possible - but no simpler.

I don't think any designer makes amplifiers more complicated because it's fun. I suspect that the more complex designs are done because over time the complexities have proven their worth. If I had to summarize what you found, that would be it.

Diffamps produce more predictable inputs. Current source loaded voltage amps have higher gain, more predictable gain, and wider swing to the supplies than bootstrap loaded VAs. Dual diffamp plus complementary mirror-image VA designs produce symmetrical slew rates and predictable drive to output stages, as well as clean, predictable recovery from overload and clipping.

It is the essence of good engineering to produce predictable, reliable results from highly variable, unreliable parts. This was an important part of circuit design teaching at one time in EE curricula. It may not be any more. I guess I'm not surprised at all that a modern, modestly complicated circuit should produce measurably better results than a more rough-and-ready simpler circuit.

The game is different for solid state guitar amps than for solid state hifi amps. SS amps readily produce extremely accurate reproductions of what their inputs say to do unless and until they clip. With as much cheap, clean power available as SS amps provide, it's easy to make much larger clean output power amps than with tubes. The trick is then to feed them an input that makes them reproduce very accurately the sound of a tube amp. It's gotta be possible - PA systems do it all the time, yes?

skey

Maybe the difference in complexity is just what measures different between amps?  Listening there's really not much of a difference.  I know I have good and bad amps all over my house and cars.  In the end I tune out any sound the amp contributes and just listen to the music.

On tube sound.  Yes any clean amp can reproduce a tube sound.  Just play a recording or mic a tube amp.  But what makes that tube amp sound in the first place?

All articles on what makes "tube" sound always show a sine wave, and squashing during clipping on the sine wave. 
for example: http://www.paia.com/tubesnd.htm

It just doesn't seem complete.

Hitting a string on a guitar is a complex waveform with harmonics.  More so on a bass guitar with round wounds.

As mentioned elsewhere, with a clean amp, when the fundamental clips, the clipping itself doesn't really add that much more harmonic content.  Instead what happens is the fundamental compresses as it can't get any louder, and the high frequency harmonics of the input signal grow stronger and stronger as they haven't clipped yet.  This unbalance might be characteristic SS harshness.

Why would a tube be better?  I just have gut feeling it has to do with roll-off of the high frequency as it approaches clipping.

Maybe - there's usually transformers in tube power amp circuits, this might help in roll off of the highs.   

Maybe it's the tube itself, when it approaches clipping for a class-a preamp, on one side of the waveform there's build up of electrons, on the other end there's starvation of electrons.  I have a feeling that at one of these extremes, high frequency can't pass.  Like at the high end of the curve, maybe a high abundance of electrons doesn't let as much highs through.  Or during starvation - high frequencies can't pass, so it may be adding to the roll-off.  I haven't been able to find frequency response curves of a tube amp as it approaches and goes into clipping.   I don't have a tube amp to measure, and I don't know how to simulate one.

I've seen that a mosfet amp can be made with the exact same squashing behavior of a sine wave as a tube amp, but I don't think it's going to be the same HF roll-off with a complex waveform at clipping.


teemuk

The circuits i simulated presented only a small partion of different power amp circuits. For example, I could replace the VAS stage with either a simple common emitter or with a complementary feedback pair. Both would create different type of clipping thus changing the amplifier tone. I could convert to one rail supply, start shaping the tone within the feedback loop or just change feedback topology to feedback current instead of voltage - or both.

I chose not to. Some people claim that simple circuits sound better - now this claim is quite easy for most of us to believe, especially because most of the good sounding guitar PA circuits seem to be rather simple. I just wanted to know whether this claim holds a truth or not. As can be read from the results, there were some differences but they were undetectable by (at least my) ears. I guess this would be equally true in almost any topology i could test. The reason for that good tone must come from some other design aspects and the simplicity is easily explained: It's simple just because it can be.

Tubes? I don't want to mess with them, everyone seems to have too strong opinions about it. In general, i think their sound is overrated, as is the case with fets too: The circuitry makes more difference than the tube itself. Anyway, how many people actually can distinguish the tube tone? Even my petty Zoom 606 digital multieffect has fooled some people to think that my 12W solid state amp was a tube amplifier. The reason why people prefer tube PA's tone is that the power amplifier has a transformer which does some neat tricks. An inductor as a load behaves way differently than a resistor, you know. I'm quite sure that similar results could be achieved with transformer coupled high voltage fet drivers. Of course tube fanatics would disagree about anything with silicon sounding like valves, even when the hard eveidence would show that they're wrong. This is an endless debate and i rather stay out of it.

Anyway, since my test didn't seem to produce any wild results i think i might try to expand it and convert the simple topology to a "vintage" topology with transformer coupling. I probably will test it with only drivers coupled and with both drivers and the load coupled. That is, when i have time.

Stompin_Tom

Cool. Thanks for the explanation teemuk. I'm excited to see if you simulations using a output transformer(?)... seems this was a topic not to long ago over at the stompboxes page... It seems to be down at the moment, but I'll post a link to the thread as soon as I can find it.

Seems like you could simulate the high end squishing and, to some degree, the power amp sag of a tube amp with a compressor... or a couple of compressors (one before and one after the preamp?).

teemuk

#10
I expanded the simulation to compare between transformer coupled circuits...

The direct coupling of the voltage amplifier stage can be replaced with an interstage push-pull configuration as follows: Direct coupled load on collector is replaced with the primary of the interstage transformer. Gates are biased with a network of four resistors and secondaries control gate voltages in opposite polarity. PNP transistor "side" can be replaced with another NPN side. "Bootstrapping" is done by bypassing one gate to the output and the other gate to Vee. Here's one example of the output stage configuration:

http://www.univox.org/pics/schematics/pa960_column.gif

1. As expected, even order harmonics decrease due to push-pull drive. However, odd harmonics increase since the circuit is not as linear as the "modern" circuit.
2. As expected, the transformer forms a high- and low-pass filter which decreases both low and high frequency response.
3. Voltage swing ability is weakened and the circuit needs higher supply voltages to match up the "modern" circuit. For the same reason the circuit will also clip earlier. The described bootstrap circuitry helps a lot but is still insufficient.
4. The supply voltages of an interstage coupled circuit "sag" much more.
5. Between 20-80 Hz the interstage coupled circuit delivers more power than the "modern" one.
6. Although differences were much more clearly visible when waveforms were compared i couldn't detect any of them by listening.

Things really change when also the output is transformer coupled. First of all, the circuit after the voltage amplifier stage is no more a "buffer" like in other power amplifier topologies so far but also provides voltage gain since power transistors operate in a common emitter configuration seeing primary inductance as collector load. See example of configuration in here:

http://www.vintage-radio.com/images/figs/transistor/class-b-output-1.gif

As expected, the collector loading reflects the impedance of the speaker meaning that the output stage no longer operates at a constant voltage gain: More power is fed to high impedance loads and less to low. When P = U^2 / R or I^2 * R, a current driven circuit will put out more power with a  larger value of R - assuming it is provided with enough voltage, which is also the major drawback of current drive. As a result the amplifier is "less powerful" on small loads but feeds a lot more power to, for example, speaker in resonance. This is the "punch" that solid state amps are reputed to lack.

Here's few things that i noticed:
1. As expected, the transformer forms a low-pass filter and highs roll-off faster.
2. Surprisingly, the "modern" circuit has a much weaker low frequency response and the bass begins to roll-off much steeper.
3. The frequency response of the "modern" circuit is quite flat when compared to transformer coupled circuit which shows more gain on higher speaker impedances. If voltage conditions are met, the amplifier is definitely more powerful on speaker's resonance and higher freqs. There was a figure showing the same result in W. Stephen Bussey's and Robert M. Haigler's article Tubes versus Transistors in Electric Guitar Amplifiers.

http://milbert.com/articles/TvsT/tvtiega.html

As we can see, this is not actually a valve character. Also, there are ways to get the same results on "modern" circuits but that's another story. For those who are interested about it see Rod Elliott´s excellent article about variable amplifier impedances:

http://sound.westhost.com/project56.htm

The tone of both amplifiers is indeed very different and it's definitely very easy to distinct the tones from each other. For the sake of truth i should say that neither one of the circuits reproducts the original signal perfectly: The "modern" can not provide enough power for high impedance loads and the "vintage" one provides too much of it. The power vs. frequency plot is not flat on either one.

A word of warning: The transformer coupled circuits are highly more unstabile and a lot more "unpredictable". The result of having "wrong" component values may not be just distortion, it can also be crazy oscillations or very high voltage peaks. These circuits are definitely much more difficult to design and tweak.

teemuk

Here's some plots. Note that the impedance would vary according to specific speaker load. A speaker model with a free air resonance peak @ 88Hz was used in these simulations.

joecool85

Why does each one get two lines?  IE - Vintage has two lines on each graph.
Life is what you make it.
Still rockin' the Dean Markley K-20X
thatraymond.com

teemuk

The other line is the phase shift. There's an error in naming of the plots too that might cause confusion: The second plots does not represent a function V(x) but a function of V(x)*I(Z), which is power.

RDV

It seems that you(I) are not allowed to download or view attachments on this board.