Amplifier testing

RobHolt

RobHolt

Moderator
Joined
Oct 23, 2004
Messages
4,126
Reaction score
12
Participating in one of the annual 'amplifiers sound the same' discussion and poll threads on another forum got me thinking about what actually matters when measuring amplifiers and what would constitute a safe, basic spec.

One thing that came to mind is that nearly every manufacturer gives distortion specs for performance at 20kHz and that labs often test at this frequency.
I don't know how many people have tried listening to music with everything below say 15kHz sharply filtered, but there isn't a great deal left!
Add to this the fact that most people cant hear a 20kHz signal, and of those who can, discrimination is extremely poor with levels reduced, and you start to wonder why more emphasis isn't place on testing related to things we can hear and where distortions are far more audible.

I guess there is always a case for painting the worst possible picture but it should surely have some relevance to reality. I'd suggest that performance at 15kHz or even 10k is far more important for assessing amplifier competency.

Essentially I'm saying that distortions become less bothersome as you approach the top of the audible band, unless they are truly bad.

What do others think?
Please also suggest what you regard as a solid basic spec for amplifiers.
I have one, but you go first :)
 
Not sure how long I can stand ordering my thoughts and typing.. but here goes, a short list of points to provoke discussion:

1) Agree that most anything 'measured' at 20Khz is of marginal value unless really exceptional instrumentation is available. I'd include measure %thd here, not because the idea is without merit - it is, very definitely - but because most measurement systems only have a bandwidth of 40-50Khz, so the result is truly misleading. THD measured at 20Khz with even 80Khz bandwidth only inculde two harmonics.. . you can see what I mean. Same goes for signal:noise and so forth BTW. So I'd rather see results normalised at say 1-3Khz, where the ear is most sensitive (and still a decade or two short of the measurement limit). The shape of the 'noise floor' is instructive though IMO, if the measurement bandwidth is known/stated.

2) Amp output Z versus frequency. You never see this, but it makes for a very informed idea how an amp will sound with any speaker...for even the most hardened cynic ;) Lots of amps measure well under only 100Hz or so, after which it is a crap shoot.

3)PSRR vs Frequency. Really hard to measure in a satisfactory way in-situ, and will effectively vary with load. One of those hidden subtleties that dominates 'good sound' I think.

4) Rated power - actually try this sometime, it's really goddamn easy to blow things up! Many amps will make 'a number' under unrealistic assumptions of mains supply, and a resistive load etc, but try plugging into the wall and using a representative length of speaker cable into a more represntative load..it's not pretty. One of thoes things wher 'keep it simple, stupid' works (also one reaosn I think Naim have the following they do - the amps are simple, but very well sorted, and will deliver what is on the rating, probably unlike some others)

By way of contrast here is an allegedly 'very fragile' amp delivering plenty:
dpa_fix.jpg


It's one of those dodgy old Deltec amps (no, not actually my own) bashing over 65V pk-pk at 2Khz into 4R7 + 100nF...after 30mins ;) Not bad for a 50w/8ohm amp.
 
Last edited by a moderator:
Martin, I've heard of conservative amplifier ratings but that takes the biscuit!

The amp output z versus frequency interests me.
NFB and the effects of any output indictor are going to figure in the results for most amplifiers, bus as you say this is hardly ever mentioned.

IIRC Paul Miller measures this if you download the full test reports from Hi-Fi News. Well worth looking at those reports as what appears in the mag are only the headlines.

When posing the question I had in mind some of the research into our ability to differentiate small changes in SPL, pitch, distortion of all types and out ultimate ability to hear low level sounds at all. In other words, where we to write a spec for the human ear it wouldn't look very impressive.
So while we can plough-on refining the performance of electronics I think we passed the point where it matters many years ago.

These days, progress has more to do with ' because we can' rather than any real need.
 
Yes I'm curious the reason why IMD is often measured with 19+20KHz? Why not 1+2KHz?

Since the 2nd harmonic of 10KHz is 20KHz, I'd suggest harmonic distortion specs above 10KHz are irrelevant anyway.

Having been designing an amp recently, I've a couple of times listened to it and thought it was working just fine then measured it and found performance way out of ideal spec. Just goes to show how low the distortion of a good working amp is compared to what we can actually hear. As you say though, make it better because we can, and why not!
 
By way of contrast here is an allegedly 'very fragile' amp delivering plenty:p.
It's one of those dodgy old Deltec amps (no, not actually my own) bashing over 65V pk-pk at 2Khz into 4R7 + 100nF...after 30mins Not bad for a 50w/8ohm amp.
Click to expand...
but your only driving 1 channel,load up both channels and see what the results are like.
 
Good point. In this case the result is the same though - clipping level is set by the ( regulated) rails of the input/Vas stage. Gets a bit warm though ;)
 
Perhaps then there is some merit in using op amps with plenty of gain for the power amp input stages. Should ensure superb PSRR for the early stages.

I recall reading an article recently where it was suggested that the basis of the Quad 405 adea centred on adding feedforward and current dumpers around an op amp, given you can get very low distortion from them. Didn't quite end up like that of course due to limitation with maximum voltage allowed and the performance from the chips around in the 1970s.
 
Yes; and if you deal with that (you can either bootstrap the opamp supply rails or use a supplemntary VAs or an output stage with gain) you end up with other problems, mostly revolving around the very limited allowable dissipation in the opamp severely limiting the 'drive' current available for what follows. So you need another intermediate buffer...After which - good luck sorting-out the compensation mis-match between a really 'fast' front end and a sluggish (max 1/10 - 20th the Ft) power buffer (hint, becomes more than one kloop to close) And the PSRR advantage mostly disappears in the detail, too. It's one of those ideas that's great on paper, but probably tends not to work as well as traditional solutions. Unless you only wanted a 10w amp when the result can be superb.

Personal opinion - the opamp front end appraoch can work well if either your aims are 'good enough at the price' * or, like certain Pro amps you can just throw hardware at the problem. An interesting intermediate may be to use a power IC opamp and add-on a beefy output stage [...long time since I played with that, and I didn't do it well then, so I can't really add to that]

Incidentally Deltec did it the 'brute-force' way - design and fab yer own hybrid opamp front end that runs on high rails, with enough output current to drive the chosen OPS, and deal with the complex 'compensation' / feedback task as part of that design problem. Not a universal solution by any means.

*e.g Rega Brio1/2 -uses an OPA604 bossing TIP31C/32C outputs at a gain of 2 or 3x about, easy and effective and robust if a tad crude. But it is bulletproof; the opamp runs out of drive current before the output devices will blow, even into a dead short. The result is the perfect student party amp, you cannot blow it up, despite no SOA circuitry paid-for; although there's a small fuse onboard for 'early warning' IIRC.

Or take most power opamps, which have nowhere near equivalent SOA of the 'Brio' and some odd thermal problems of their own - and often some very invasive 'protection' circuitry to cope as a result.
 
Last edited by a moderator:
This TdP circuit sounds rather nice, and it can work perfectly well with just one pair of output devices for an even simpler circuit:

mf_p-amp_170_scaled.jpg
 
Opamp, buffer stage with a gain of roughly 3x, driving mosfets - yup, that'll work ;)

The devil is in all those little caps to strap it down; and the 'bias' setting arangement is , um, rudimentary. And making a clone using a 'modern' opamp would be a whole new bundle of fun :D
 
I got to play with one recently (MF P170).

Nice sound and worked well, though the worst build I've ever seen in a commercial amp. I mean, circuit boards secured by the output transistor lead-outs!
Not to mention the scorch marks under the op amp rail dropper resistors.
 
You can achieve very low noise and also significant current drive by using many parallel 5532's as per the 'ultra-low noise input stage' of Doug Self. Not only is the noise reduced by the parallel chips, but with more output current you can run higher voltages at output too.
 
But it's only half the problem cracked (incidentally, Jan Meier sold poweramps based on 50+opamps in parallel over 10yrs ago - LM6172s IIRC).

One thing I'm still playing-with is trying to just use the opamp for purely error correction... must get this moving again...
 
Oh yeah! I saw that amp on the website when I bought my AKG K1000 from him. I thought it was an interesting idea but figured it wouldn't really work that well 'cos it's so odd-ball.

I find using 5532 op-amps in series seems to really up IMD though I'm not sure why it does this more than THD.

Anyway we haven't arrived at a spec for a good amp! Are we talking about what we consider reasonable today, or what is audible?
 
Ok, for starters:

THD <0.05% up to rated output and up to 10kHz

THD+N <0.1%

IMD <0.05% as above

Outout impedance <0.3 Ohms below 100Hz and <0.6 Ohms at 10kHz.

Channel sep better than 40dB

Frequency response deviation less than 0.25dB between 20Hz-15kHz
+/- 0.5dB between 20Hz-20kHz.

Figures to be maintained with input signal meeting amplifier specifications and when output is loaded within specification. Line inputs only.
Also presumes a nominal 8 Ohm load. No 2 Ohm ribbons!
 
I'd have to make a more strict spec for something truly in-audible. Mostly I'd insist that IMD is lower, because this is about what the DEQ has in standard guise and while not obvious on first listen, does add some hash that you can hear on a good system. I'd put this down to the very wide spread that can occur with real music inputs.

The THD at this kind of level would need a further restriction. Firstly I'd say at 80% of rated output, because full rated output usually rises a fair bit on most amps (or they should rate more conservatively) and secondly that each harmonic is at least 6dB below the last as order increases.

THD <0.05% up to 80% of rated output and up to 10kHz

THD+N <0.1%

IMD <0.01% as above

Outout impedance <0.3 Ohms below 100Hz and <0.6 Ohms at 10kHz.

Channel sep better than 40dB

Frequency response deviation less than 0.25dB between 20Hz-15kHz
+/- 0.5dB between 20Hz-20kHz.
 
IMD tends to track THD.

Look at amplifiers with good THD and you'll see good IMD, certainly from a glance at the figures for around 20 integrated amps would confirm it for linear amplifiers.

My figures are a base spec and I doubt most people could detect them. In reality you would build in some margin.

I'm also not saying that these distortions are in themselves going to make a poor or below average sounding amplifier. Not all distortions sound obviously bad and some can be subjectively preferred. Just looking at a difference threshold.
 

Latest posts

Back
Top