Another Discrete Amp - GainBloak

GroupDIY Audio Forum

Help Support GroupDIY Audio Forum:

This site may earn a commission from merchant affiliate links, including eBay, Amazon, and others.

tk@halmi

Well-known member
Joined
Jun 3, 2004
Messages
999
Location
Oregon, USA
I am soliciting criticism of this circuit. :grin: My aim was to create a small amp for DIY use that:
- Uses easy to obtain BJTs
- Does not require matching of transistors
- Has single ended power output
- Has simple and predictable gain setting

Circuit

It lives in simulation only for now. This weekend I may have the time to breadboard it.

Cheers,
Tamas
 
Just some random thoughts:

Have you also tried transient simulations ?

How about stability, does it ring ? Gain is high (40dB) so more easy there.

Is the difference in DC between in & out OK for you applications or for an universal amp-bloke ? (shouldn't be too big a problem)
 
[quote author="clintrubber"]Just some random thoughts:
Have you also tried transient simulations ?
How about stability, does it ring ? Gain is high (40dB) so more easy there.
Is the difference in DC between in & out OK for you applications or for an universal amp-bloke ? (shouldn't be too big a problem)[/quote]

I do time domain transients and AC sweeps all the time.
In the simulation it was unconditionally stable at a gain of 2.
However, simulators are not to be trusted and need to be confirmed
by a protoype. The gain-bandwidth is suspiciously constant over a
wide range of gains. Perhaps Orcad Pspice is kidding me.

Yes, capacitor is required at the output as it is always at -0.7 volt or so.

There are always trade-offs in a design. Here we traded input pair
matching with DC offset on the output.
I tried to employ mechanisms that avoid the generation of odd
harmonics and cross-over distortion.

Tamas
 
It's not a bad little stage in terms of performance for using 5 transistors imo.

The only real thing I dislike is it can clip way ugly, if the output runs out of current. It is more or less the consequence of the asymmetry of the output stage, and happens when it is loaded too much. In your circuit, with the high closed-loop gain and light loading by the feedback network, you are pretty much ok, since the Q20 current source will pull the 600 ohm load down adequately. As you go to smaller gain values and if you keep R113 small you get into trouble, and there are reasons to keep the feedback divider impedance low.

I would run the diode bias a little richer---with the current lavished on the output stage might as well anchor the bases of Q18 and Q20 firmly.

I don't see any advantage to having R124 in the emitter of Q19, but no great disadvantage either. It could be in the collector if its purpose is to limit short-circuit current.

The bandwidth behavior is not that odd when you consider that you have in essence a current-feedback topology, with the feedback applied to the relatively low impedance emitter of Q9.

You can improve the distortion performace markedly and enhance the already healthy bandwidth by interposing a PNP emitter follower driver between the Q10-Q18 collectors' junction and the base of Q19. A pullup resistor for the new PNP emitter of about 3k is roughly optimal for distortion---at least based on a simulation with a closed loop gain of 5 (R113 = 200 ohms, R114 = 800 ohms). Under these conditions the -3dB bandwidth is about 10MHz and the THD for 5V p-p out is only predicted to be 2.6 ppm @ 1kHz. Undoubtedly signal-induced self-heating will spoil this a fair bit---my program does not allow an easy modelling of those effects, which are significant at such otherwise low nonlinearities.

You mention readily available transistors---I know what you mean, but there are sources albeit less convenient than the majors like digikey for the really fine japanese bipolars and (sadly dwindling/discontinued) JFETs. The far east uses small signal bipolars by the truckload that are imo manifestly superior to most readily available ones in the states. They cost about 1.3 cents in those quantities as well.

Brad
 
> The gain-bandwidth is suspiciously constant over a wide range of gains. Perhaps Orcad Pspice is kidding me.

SPICE lies a lot. Or to be fair: it often gives the right answer for the unreal world that SPICE lives in.

But if you are changing gain by changing R113: yes, gain-bandwidth will be roughly constant until R113 is less than a few ohms. You have re-invented "current mode feedback".

This thing eats a LOT of power.

> I don't see any advantage to having R124 in the emitter of Q19

As shown, nearly none.

Also if the feedback network were lower-Z, R23 is a 2-cent waste.

> interposing a PNP emitter follower driver

If you do that, R124 12Ω may have a small advantage. It linearizes Q19's Gm. That does not matter when Q19 is driven from a very high-Z source (naked collectors of Q10 Q18), but starts to come into play with a buffer. I'm not sure it makes any real-world difference.

In general, numeric THD will drop if R27 is made large, so Q10 base current is around half of Q9's collector current. This defies conservative design, and I'm not sure the simulated improvement is real.

Brad: it is good to have another savvy circuit-reader on the forum. Welcome.
 
PRR: Thanks---pleased to meet you as well. How do you have the time for all these posts (he asks himself as well here)?

Actually I think there is a fortuitous cancellation effect going on with the added PNP, involving both the nonlinearity in Q19 driving 600 ohms, and even extending to the high Z drive before it. Simulation showed a lower distortion overall than when a "perfect" op amp follower was substituted (infinite input Z, 100dB flat gain out to light---circuitmaker's OPAMP5 model). This was unexpected. Not to recommend for production to say the least, but the value of the pullup resistor on the emitter had a broad optimal region for minimum distortion.

I remember the fully complementary emitter follower design being touted by Rich May (ex. of Sumo, then with Harman*) as good for audio years ago; I believe he attributed it to HP at the time. That is, the P's driving N's and N's driving P's, with a rough cancellation of net offset voltages. He added that if you bootstrapped the collectors of the input devices to take out the voltage-dependent Ccb's it was a very accurate stage. I shall have to do some more quantification of the performance.

Brad

*I believe a high-end electronic crossover box was actually built and sold before the management types characteristic of the company decided someone was having way too much fun. Ooops---I hear the late Jack Webb saying "Only the facts, ma'am"
 
Thank You Brad and PRR for the feedback.

I will try adding the emitter follower recommended. My main concern with increased OLG is preserving stability at low gain settings.

I am not happy with the clipping characteristic myself. When loaded mildly it clips asymmetrically. Isn't that intrinsic to the single ended output configuration? If you load it more it turns really ugly and choppy. So you have to run a lot of current running throug it to avoid that. If I am going to waste 50ma in the output stage should I just go with a complemetary follower output?

Also, the circuit likes to run on higher voltage rails than I initially thought. The performance seems to be better at +-24V than the 20V shown originally.

In the United States we are a little spoiled with the availability of parts from all over the world. We tend to think that everyone has the same access to this plentiness as well. If you ask builders from South America or DIY folks on the edges of Europe they will tell you that getting just common Digikey parts can be a difficult task.

Cheers,
Tamas
 
Tamas, I would favor complementary unless you really like to have second-order distortion dominate (it will anyway because of the single-ended input, so why worry I guess). As the bartender says in the movie Harvey when asked about the reality of the large rabbit, "Well, there are two schools of thought..." Of course you can run the complementary stage as rich as you like with adequate emitter ballasting and heatsinks, if you like to burn the power.

But in this direction of constant improvement lies madness, as better and better performance is pursued and the transistors multiply. The really cool output buffer I mentioned above takes four, six if you run the two input devices from current sources. It is fast though. Then, with a buffer that good, one gets pushed back to look at the preceding stages. Pretty soon you've got a complex beast that achieves the performance of a power current-feedback amp, with some possible advantages like less concern of the output stages heating up the input stages and inducing distortion at low frequencies, as well as potentially lower noise.

You could preserve your class-A approach shown and figure out some clever ways to clip symmetrically when you get close to running out of current. One of the problems is the big cap to ground: it allows you to get away with high audio gain with only the one Vbe of offset, but when you overload you have time to rearrange your sock drawer.

BTW I think the added follower I suggested may actually improve stability at low gains owing to making the collector-base C of Q19 less important---in any case simulations predict a textbook-perfect response down to nearly unity gain (you won't get exactly unity with R114 > 0 as R23 functions as part of the feedback divider). I would not let R144 get too small or in the event of some accident Q9 could sustain some damage. I used 800 (806 for nearest standard 1%, or 820 for 5%) ohms.

Brad
 
> When loaded mildly it clips asymmetrically.

So?

1) Who clips?

2) Who wants symmetrical clipping?

3) Isn't the clip-point WAY higher than any sane person needs?
 
It turned out to be the input transistor clipping. Reducing the preset emitter current through R23 has improved positive swing surprisingly.

I have added the follower stage between the gain stage and the output as Brad recommended and increased the current through the diode bias. People who like fancy things could use an LED instead of the diodes.
I decided that the role of R124 will be current limiting only so it goes on the collector for now.

Latest version

The simulated amp can swing +-16 Volts easily into 600 ohms now. That is good enough for me so I stopped polishing the cannonball. This weekend I will build a breadboard prototype.

Thanks Everyone for the contributions!!!
Tamas
 
I have built this thing and it works pretty close to the simulation results.
It has a large bandwidth even at 40dB gain. It is nice and stable at the x2 gain setting and the step response looks more constant over the gain range than voltage feedback amplifiers do.

The output transistors run VERY hot. There is over 70mA current running
through them using the 10 ohm resistor in the emitter of the current source. Mounting the diodes right onto that transistor will be necessary. I think a 15 ohm resistor would be better. On the protprype I ended up using a 20 ohm one because I have no way of easily mount the diodes on the transistor.

I am trying to decide if I should go with the complementary output instead of the single ended one. Most of the power seems to go to heating the room on the single ended output instead of driving the load.

Any suggestions?

Tamas
 
Well, of course that's the Way of class A.... At least you don't have output stage crossover distortion. Actually I don't know that you need thermal feedback as long as the current source has ambient compensation, which it does with the two-diode bias network---just make sure the heatsinks are big enough. Thermal feedback should be avoided imo if you don't need it.

If you go to a complementary output stage then you may need some thermal feedback, depending on the emitter resistors. The topology I would recommend is the PNP driving the output NPN and a new NPN driving the new output PNP. If you will send me your email I'll send you a file showing what I am talking about if that description isn't enough.

Do you have any way of measuring your distortion performance? I'm intrigued to know how close it comes to simulation results. And of course, how does it sound?

Brad
 
Hi Brad,

After some more testing I think I am staying with the single ended output. It is very simple and all there is to adjust is the current using that emitter resistor. Rearranged the breadboard layout and mounted some heatsinks to the transistors so they can be run and tested at high currents.
Thanks for the advice on the thermal feedback. I thought it would help, but in case of a dead short on the output it probably does not matter.

Performance that I can observe on the oscilloscope is outstanding. Bandwidth, slewrate and gain are excellent. At 2KHz things look perfect. There appers to be a tiny bit of overshoot followeed by a little ringing on the square wave at 20KHz and above. This is mostly at high gains, from 35 to 40dB.


For distortion measurements I use Right Mark Audio Analyzer that runs on the PC. It gets down to 0.0003% distortion on sunny days. Today I will make some measurements.

Listening tests may have to wait another week.

Thank You,
Tamas
 
> I am trying to decide if I should go with the complementary output instead of the single ended one. Most of the power seems to go to heating the room on the single ended output instead of driving the load.

> Well, of course that's the Way of class A....

Yes and no.

What we have here is one-sided class A. To get 70mA peak load current, we use a 70mA current source. That forces the active device to swing 0mA-140mA. This is hard work, and not necessarily low distortion.

Maximum efficiency of this plan is 25%.

Actually it looks like you are asking 30mA peak load current, using a 70mA current source so the active device only has to swing 40mA-100mA, not a great variation of Gm. But max eff is below 12%.

We don't have true current sources. We have current limiters that have to be fed excess voltage. You could cut power waste in half with a true current source. You can come pretty close, over the audio band, with a good coil (though that loses the convenient bias-setting). With a coil, max efficiency is 50%, though often more like 33% or 25% to reduce GM variation and THD.

Now if you can drive two devices in true push-pull, you can get back to 50% efficiency and also have a first-order cancellation of GM swing and THD. If you want 30mA peak load current, idle the devices at 15mA, swinging 0mA to 30mA to drive the load. Or maybe 20mA idle, swinging 5mA-35mA. Gm will vary from (expressed as dynamic emitter resistance) 1.5Ω idle, 6Ω to 0.8Ω swing. This is a total output resistance of 0.75Ω idle, 0.7Ω at peak swing, nice and constant. (Adding emitter resistors reduces the percent change, but not the absolute variation of output resistance.)

But if you have come this far, it is worth looking at class B. Or really: class A for low outputs or low load resistances, class B to carry big outputs in low loads. Most of your loads are really 2K or even 10K. And you rarely swing 15V peaks, and might rarely hear distortion if it only happens on peaks. 10V peak in 10K is 1mA peak, idle the pair at 0.5mA. This cuts power demand to 1% of your pull-only over-biased class A output. You may still demand 30mA peaks in 600Ω, which leads to a series of compromises with current swing, bias resistors, and thermal stability. But loudspeaker amplifiers are often worked 50mA idle 3A peaks, the same ratio, and they can sound good and be thermally stable when done right. Or for this small-scale low-compromise project, you might aim at 5mA-10mA idle, still a LOT less heat and power-supply than the 70mA pull-only output.
 
Thanks PRR. Perhaps I will make a push-pull output version later. I am sure someone will want to hook it up a certain 75 ohm transformer to it :roll:, and that will not happen with a single ended output.

I got some THD figures. This is with 470 ohm resistor load.
The output never exceeded 0dB because RMAA cannot deal with that and I haven't put a pad to the output yet.

@1KHz:
~6dB gain - 0.0004
~24dB gain - 0.0004

@7Khz:
~6dB gain - 0.0007
~24dB gain - 0.0008

40dB gain measurement has not been done yet, but planning to.

I will add a pad to the output and re-measure at 14db output after I had some pizza.

Tamas
 
[quote author="bcarso"]Kewl! So you are essentially at oscillator/analyzer residuals. It is probably even better than that. Congrats.[/quote]

Well, I could not have done it without Your and PRR's help.
I just hope that RMAA is telling the truth.

Thanks,
Tamas
 
:grin:

I just could not believe the numbers that RMAA generated. Knowing that I have never achieved better than -80 some dB noise floor on my bench top and looking at the -105dB noise reported I grew very suspicious. I have decided to review my test procedures from beginning to end.
I was not all that surprised to learn that in my excitement I was reporting measurement results in the reference channel that is simply a loopback cable. Thankfully the media has not been alerted and I got away with just some self inflicted shame. :oops:

After a few minutes of whimpering I looked at the spectrum analyzer and I could see that there was plenty of harmonic distortion present. I should have checked this in the first place.

Now I set out to make some THD measurements using RMAA, but checking for data for the channel that included the test device. The result was a nasty 0.25% distortion when using the second version of the amp. Ughhh! This was not expected at all.

When I reverted to the original amp without the follower inserted the distortion dropped back to 0.033%. This was even more of a surprise to me. Now the 2nd harmonic was down to -70dB, 3rd down to -90dB and everything else submerged in the noise floor.

With optimization to component values some improvement could be made to these figures. However, I want to see the distortion from the 0.01% to 0.005% range and that requires a change to the topology. The second version would have worked much better with a current source feeding the source follower.
However, I am determined not to add yet another transistor so I am moving the source follower from between the VAS stage and the output to right after the input transistor. Also the current will be ramped up through the VAS. This should yield far better to improve available gain according to the simulator.

Sit tight there are more interesting developements to come tomorrow night.

Tamas
 

Latest posts

Back
Top