discrete vs. IC opamps? has anyone tried the pass labs design?

[dorkus]

hi,

has anyone here tried constructing a discrete op amp design, e.g. as shown on the pass labs web site? was curious how it compares to a good IC (OPA627, AD825)... i like the simplicity of a good discrete circuit, implementing the whole op amp with just 8 transistors... in fact, Bryston equipment uses such discrete op amps in their preamps and they seem to sound quite good. any thoughts?

marc

 

[Samuel Jayaraj]

Marc,

I have made the discrete opamp from the Pass site. Since the MPSA18 suggested in the design is not available where I live, I used 2N5551 in its place. I'll describe the design as well as I can. A differential amplifier formed by 2N5551; the collector of the first transistor is connected to supply via a 820E resistor while the second transistor's collector goes directly to positive supply. Their emmiters are tied together and I used a current source formed by two 2N5551s. For the second stage I used an MPSA92 with a Miller cap of 27pf. For this stage too I mirrored the current source from the first stage current source and again mirrored one more current source for the single ended class A output stage formed by BC337-25. In the feedback loop I incorporated a 10pf cap, as well as the same value for the input shunt. The performance is extraordinary to say the least. It is very quiet compared to FET and Bipolar IC opamps. I wish I could post the full schematic, but I don't know how to. I hope you have got an idea.

 

[mlloyd1]

I have no doubt the discrete opamp described does a good job. However, I'd be VERY uncomfortable making the blanket statement that it outperforms some of the best IC opamps available (OPA625, AD825, AD797, some others). For example, I don't think the 2N5551 series is especially known for low noise. Even for a low noise device, there is often a range of optimum bias currents for best noise performance from a given source impedance. The optimum bias for lowest noise may not even correspond to optimum bias for lowest distortion or best frequency response for the application. Etc.............................

There was a similar discussion thread a short time ago (a month or two? could probably do a search) where the discussion talked about comparing the "best" discrete design to the "best" IC opamp design. In this instance, I'm very comfortable saying the best discrete is probably better overall, because the designer has control over virtually all parameters and can optimize accordingly to his/her level of skill or desire. This flexibility is not nearly as available in IC op amps, although performance is radically impacted by the surrounding circuitry (e.g. power supplies, compensation, etc.).

What's my point? You can make really good gear with most good technology technology PROPERLY applied. I have some buddies who are really into tubes. Sometime I play games with them by bringing over whatever my latest homebrew preamp is and talk about the tubes I use in it. I put it in their system and listen to them talk about how good it sounds. Then, I break down and tell them it uses some IC op amp.

Yes, they still speak to me. They just make me open any boxes I bring over now


Have fun....
Michael

 

[Samuel Jayaraj]

Michael,

I do agree to your point that the designer has control over the entire design in discrete circuitry, but IC designs offer more flexibility. Given the right bias, the 2N5551s do have very low noise and low distortion to boot. The reasons why I used these devices, not common to audio, is that I saw them used in the Stochino Ultra-Fast Amplifier design and I have made this amplifier myself and know how very good it sounds (hence, I had many surplus devices) and I couldn't find anything (read available) close to the MPSA18. I ran extensive simulations with Electronics Workbench and found that devices like BC550C etc., exhibited more distortion than the 5551. Therefore, I did not regret the choice, though I didn't have one.

I have pitched this discrete design against NE5534 (Reg Williamson's design from Electronics World entited 'Simple but Sound' with an active volume control), various NE5532 designs, an OPA604 design (a very good one on its own rights), Bride of Zen and another discrete Class A design using MAT02 and MAT03 matched devices for the input differential pair (Elektor design).

In all of the comparisons, the discrete opamp design sounded right (that is added the least of its own signature). It did not add any warmth or glow, but was not sterile either. It did sound a bit lean (I don't mean not enough body to the music) as compared to a good tube design (which makes the sound more rounded) but for the simple circuitry, I felt that its neutrality and fine detailing and resolution were indeed outstanding.

I have also checked this on simulations against the Low Memory Distortion design that I obtained through a link from here. That seems to be a better design since on a square wave test for 10mS, I find that the simple design's trace begins to move down the scope-screen whereas the latter trace remains on the same horizontal axis. This again is a fully discrete design. But with this circuit I found that it is not easy to change the feedback network since the circuit stops operating, showing errors with the FETs in the input section. (Right, about the flexibility aspect).

I do however, agree that there might be better OpAmps like the ones that you have cited or perhaps, the OPA627 etc., which might sound just as good or better.

[Edited by Samuel Jayaraj on 06-21-2001 at 08:02 AM]

 

[jam]

Op-Amps

Sam,

The leaness you describe could be alleviated somewhat by reducing the loop gain of the circuit by adding more degeneration to the differential and gain stage. Try increasing the values of the emitter resistors in the differential and the gain(VAS) stage.

The NE5534 and NE5532 are pretty old designs and some of the newer op-amps Michael mentions sound much better.
In my experiments I have found that adding a buffer to the op-amp inside the feedback loop usually always helps.

Jam

 

[mlloyd1]

Marc asks:
.... Bryston equipment uses such discrete op amps in their preamps and they seem to sound quite good. any thoughts?

I've heard some Bryston preamps and power amps. They DO sound great. I've read in the Audio Critic (Dr. Rich's writings are great reading for anyone interested in audio DIY!) that Bryston doesn't use fancy things like current mirrors, etc. in their designs (although the power amps output stages use a clever "trick" or two); they keep things simple to get the extremely high reliability. They get very high performance from such simple circuits with years of experience of optimization: refining device choices and bias points, refining layout and wiring, etc. I'd guess that it would probably be tough to duplicate their performance if they just gave you a schematic because of all of the optimization.

Bottom line: be prepared (i.e. have patience and be willing to learn) and equipped (i.e. with databooks, o'scope, access to other useful test gear, etc.) to tweak to your heart's (and ears'!) content

Michael

 

[dorkus]

i agree with what's been said here... IC opamps definitely have the edge in implementation simplicity and flexibility. being lazy, i am leaning towards going the IC route just because the results are much more "guaranteed" than with discrete design. but if you're willing to tweak, discrete definitely has a lot of potential... the extra components in monolithic IC designs (input clamping, current limiting, etc) does make them easy to use and reliable but from a purist audio standpoint they are unnecessary and better left out.

i guess the real question is... how much tweaking does it take to get a discrete design to sound better than, say, OPA627 or AD825? are there ready-made discrete implementations out there (e.g. pass labs examples) that if properly implemented will beat the aforementioned ICs in sound quality, or will i need to do a lot of fussing? i guess i am looking for an easy way out, but i'll just have to break down and do the experimentation... *sigh*...

incidentally, once i settle on what circuit to use, i am going to try to go all-surface-mount with the layout to absolutely minimize lead lengths and parasitics...

 

[Hoffmeyer]

I too was intrigued by this article about DIY-opamp's, and started on what has become a long journey.
My first idea was to make a circuit much like the one in the article, combined with the ideas in another of NP's articles, the cascode amp design.
My second idea was to make it without global feedback!!!
The third idea was to make it fully ballanced, and still with the cascode design, since I'm planing to make a ballanced poweramp (sometime in the future).
Right now I'm actually having it on the testbench, to see how it meassures, and the next natural step will be to listen to it.
Need I say that it doesn't look like my startingpoint at all? That's what I love about DIY. You never know where it might bring you.
Simulations in Orcad PSpice indicates a linear frequency response up to @15MHz, wich, off course, will be reduced at the input. +/-10 Volt out in a 5kohm load is no problem, 2.order distortion at -90dB, and 3.order even lower, but all that is still only simulated.

 

[jam]

Marc,

A long time ago I built a discrete version of an op-amp using the data book of the op-amp. They had listed all the values of the components in the device. Well the discrete version sounded better but the op-amp had a lot lower offset and the distortion was lower as well.

Your idea of using surface mount components has a lot of merit, you could get the best of both worlds, short lead lengths, lower noise pickup and the possibility better thermal tracking.

I believe that rapid heating and cooling of the output devices inside the op-amp cause some form of thermal distortion, and a buffer cures this problem by reducing the load on the output devices inside the op-amp. In fact this is what Levinson implement in their pre-amps.

But the bad news (or good) is that none of the op-amp designs that I have built is better than a good discrete design, like the Pass for example. That being said some of the newer op-amps come pretty close and it might be only a matter of time before ....

Jam

 

[ppl]

Having exparamented alot with Both I wold say in a nutshell An IC OPAMP/Buffer Combo Properly implimented with the Right Choice of Devices can Surpass a simple All Transistor Design. Also like pointed out above The Possibility of obtaining Great sount rather quickly is attractive. Space is another Consideration, Again IC'S Excell. A Transistor aproch will allow the Most Optimum solution but be Prepaired to do alot of Tweeking. and use a Quality Topology Like Casscode JFET input stage and Second Voltage gain stage eather Jfet or BJT. I think a Jfet sounds Better in the Second Stage. Better yer a Full Complimentry Input to output Topology using 2Sk389 & 2Sj109 Dual monolithic Supper matched JFETS.Curent source everything and Tweek each and every part in the Circuit then you can exceed the Performance of an opamp/Buffer Combo. Now IC's do have some drawbacks mostly lack of Dynamics. This i think is because of the limited supply voltages you can operate the Devices. I have noticed that the same circuit Eather IC or transistor sound more dynamic operating at a higher supply voltage than thay do operating on lower ones Like + & - 24 volts rather than +/- 18 Max of most opamps. Each Type of device both a single transistor and an IC opamp will have Diffent sonic presentation Example the OPA-627 sounds great so will the AD-825 and AD-797 but Each are worlds apart sound wise from smooth and Easy to listen to to Sharp and Anilytical. Also a 2N5551/2N5400? sounds way different than say a MPS8099/MPS8599.

 

[ergo]

One good place to look for information and ideas is

http://www.borbelyaudio.com

Mr. Borbely is a big fan of FETs and has many kits available online. I myself have not built any of these...
But if anyone has comments, it's alvays interesting to read them.

 

[ppl]

Borberly is Right up their with John Curl, Walt Jung , and Nelson Pass just to name a few, as one of the Classic Audio designers. An interesting point about these guy's is the fact that thay all Eather worked for IC manufactures or have learned alot of their Knolage from IC Data Books. I Know i have learned alot from IC Data Books and APPlication Notes. I have been playing around with an All transistor Design using a Input stage topology like Borberly uses and is Full Complimentry as per Mark Levenson and John Curl. Uses Casscode Stages Throught inCluding the Driver and output stages. The Proto-Type has worked and sounded Great the First time. Most of my Audiophile Frends Thought the sound was way Better than my IC Designs and these same people thought my ic designs were some of the Best sounding Line stages thay have Heard. But thought the Prototype i just discribed had Better sound stage more Delineation of detail and A term i have not heared nor could thay explane to me what thay ment "Space". I thought that indeed while the Prototype Circuit sounded better The Differencewas not as Great as I expected it to be and the IC design fit on a board 1/4 the area of the Discreet design, only took a few days to Tweek. andworked rather as expected right after The Circuit was Powered up. The Discreet component circuit has been in the Tweeking stage for about 1 year.It also worked somewhat as expected at first except i had to lower the Open-Loop gain by using Larger Source Resistors on the Input stages, This cured an oscilation problem with some loads. So in conclusion I think that if you want the Last word in Sonic quality then A discreet component circuit is the way to go but expect it to be complex create Headaces and require alot of effort to get compleatly right. on the other hand if you Know what kind of sound you like and want to get the Circuit up and Running in a short time and have space limitations then A good IC circuit is hard to beat.

 

[Samuel Jayaraj]

Jam,

Thanks for your tips on alleviating the leanness part. I think I'll try that out sometime. What I will do tonight is to get this little gain block to work off a proper power supply; actually, I ran it off a lower than intended 32 volt dual supply, the boards had no supply decoupling caps whatsoever. The reason is that I made these as modules, so that I could solder a few pins and stick the board into an IC socket for a direct comparison - IC in and next Discreet module in; this test was to be done in a crossover unit but I never got to actually doing this. The crossover board with the IC sockets would have contained the decoupling caps anyway, and if correctly rated, it was only a matter of switching over to the correct voltage in each case. Given the right supply and layout, I expect this to sound great.

As already mentioned, the only good comparisons I had at hand were the Bride of Zen, OPA604 and MAT02,03s. I cannot find better opamps in my country (India) and my only source would be RS Components.

The Electronic Workbench simulations show the following data: Differential pair run with a current source of 7mA. Second stage (MPSA92) run at 3.6mA and output stage run at 2.95mA. This is for configurations with a gain greater than unity. I found that for unity gain (Buffer) the current sources have almost got to be reversed in their order to obtain the best figures.

At a gain of 4 times, the frequency response is flat upto 200KHz. Second harmonic distortion is 0.00002% upto 5 KHz rising upto 0.000045% at 20KHz and still 0.0001% at 40KHz. Third harmonic distortion is 5 decimal places down all the way to 40KHz.

I know these are ideal figures, but the actual model seems very good too.

At a gain of ten, the frequency response remains the same. Second harmonic distortion is 0.0005% upto 5KHz and 0.001% at 20KHz. These figures look quite realistic although still ideal.

Just give me a hint as to how to upload circuit diagrams onto the web and I'll probably try and put this up. This is no claim to being the best or anything of that sort. But it is a good point to begin with, especially if the purse doesn't allow for more.

If one can try out the Low Memory Distortion type of discreet model, the results sonically could be astounding, compared to even the best 'Audio' ICs available.

 

[ppl]

Samuel While maby unable to get Good Opamps in your area you arev truly blessed to be able to still get the MAT-02 & Mat-03's I have a supply of these left over from years ago but i have been unable to locate any hear in the US nowdays. The closest i have came to these is Linear Intergrated Systems LS-312 & LS-352 complimentry Low noise Monolithic Dual PNP & NPN. BTW These work great also. Please let us Know How your Comparrison works out.

 

[Hoffmeyer]

ppl

I'm surprised that You're unable to find the MAT-02 and MAT-03 from Analog Device in the US, since they're still being manufactured!
In my own project I'm using the MAT-01 and MAT-03, since I have easy acces to them, and that's the reason for my surprise. Have You tried to find a dealer through http://www.analog.com ?

 

[ppl]

Thanks, My AD rep sed thay were Discontinued. will check with AD.

 

[dorkus]

FET or bipolar diff inputs?

hmm so are most of you using bipolar devices for the differential input of your discrete inputs, or FETs? in general i thought FET inputs are more linear, at least in IC devices. has anyone tried hybrid design, i.e. FET inputs with bipolar gain/output stages?

P.S. anyone who needs a place on the web to upload their schematics, design info, etc., please let me know - i have a web server with plenty of space and am interested in eventually setting up a nice audio site w/DIY stuff. having different people's ideas, etc. in one place would be a great start - just email me at [email protected]

 

[jam]

Fet vs. Bipolar

Marc,

The advantage the fet input device has is that the offset voltage can be very low, there is almost no gate current involved. Bi-polar input devices can have some offset due to changing base currents with different source impedences (example. volume control). So if the input on a bi-polar input device is not cap coupled the offset would change with changes of the volume control. (Since we all know there is no cap like no cap, why use one?)

A Fet input devices has the potential to have lower noise at impedences above 1k (source impedence).
I myself like fet input devices but still prefer discreet designs with mosfet inputs, even though there is a higher noise penelty. This should not be a problem unless you are designing a phono stage.

Jam

PS. Great idea about the web site, we could all pool our resources and probably come up with some nifty designs.

[Edited by jam on 06-22-2001 at 01:57 PM]

 

[dorkus]

cool. i definitely agree about the offset advantages of FET devices... so you prefer MOSFET to JFET? hmm wait isn't JFET a type of MOSFET? i forget the terminology... anyway, my mention of nonlinearity with bipolar input stages is in reference to walt jung's article on thermal distortions in opamps, but that probably applies more to monolithic IC devices where there is more thermal coupling between stages. also, the right amount of emitter degeneration on the diff devices can probably alleviate any possible nonlinearities with bipolars.

anyway, let's definitely pool information for an audio site. i'm a lead web developer at a dot-com firm (yes, we're still in business) so can help come up with a nice slick site.

marc

 

[jam]

Marc,

Well, jfets and mosfets are slightly different. I like mosfets because you can get some really low rds. devices and these seem to work better in low Z circuits.(Like bi-polars)
Bi-polars usually have a lot higher transconductance and usually some emitter degeneration helps (improves linearity).

Please let us know when you are ready, and I am sure the web site will be a big success.

Thanks,
Jam