opinions

[artu]

I want to hear opinions of this design. I've spiced it and gives me outstanding THD's at full power (note that the components are what I've found available here in Chile). I am new in this hobby, and I want to do a PCB with this circuit.

THD Frec Output amplitude Volts (8 OHM load)
0.028 400K 28.18
0.008 200K 30.20
0.0045 100K 30.78
0.0024 50K 30.93
0.0006 10K 30.98
0.0004 1K 30.98

Any comments?
Thanks in advance
Arturo

[artu]

I've added two transistors to make the voltage amp section differential. this halves the THD, stability BW, and dumping factor remains the same

[djoffe]

That's some really good results...I'm a bit concerned that your models for the output darlingtons my be a bit iffy wrt frequency response/stability. Have you sanity checked them?

[artu]

I've heard concerns about the output darlingtons, are a little bit slow adding phase delay, but in this case are hard drived from previous stages. the cascode pass thru a follower before the voltage gain stage, [the challenge was maintain stability], the phase shift added by the darlingtons is negible far above the audible region.

[artu]

Checked BDX33, the simulation with the Motorola model gives same results, attached sine wave at 100KHZ, still flat band


ngspice 3 -> fourier 100000 v(4)
Fourier analysis for v(4):
No. Harmonics: 10, THD: 0.00297656 %, Gridsize: 200, Interpolation Degree: 1

Harmonic Frequency Magnitude Phase Norm. Mag Norm. Phase
-------- --------- --------- ----- --------- -----------
0 0 0.00263235 0 0 0
1 100000 30.7938 -6.8426 1 0
2 200000 0.000257071 -93.347 8.34814e-06 -86.504
3 300000 0.000764524 113.537 2.48272e-05 120.379
4 400000 0.000238637 49.485 7.74952e-06 56.3276
5 500000 0.000299639 -54.144 9.73052e-06 -47.302
6 600000 0.000121386 -139.2 3.94191e-06 -132.36
7 700000 0.000140829 123.56 4.57331e-06 130.403
8 800000 5.71412e-05 31.1615 1.85561e-06 38.0041
9 900000 7.07213e-05 -64.973 2.29661e-06 -58.13

[CBS240]

My concern would not be the phase delay of the Darlingtons so much as the thermal compensation to keep bias stable. Keep in mind that the driver is on the same die as the output device and is also heated. So you have one temperature coefficient multiplied by the next. Complementary Darlington class AB output stages require thermal compensation....

[artu]

Good point CBS240, in fact the bias point changes, I've simulated with global temp's from 27° to 125° and the bias piont changes from -0.063 to 0.060 Volts and the bias current (class A) from 2.4A to 3.1A. I will do more acurate simulations keeping the signal circuit at one temp (35° to 65°) and the power circuit at a high temp (125° to 150°)

[artu]

I did some thermal simulations. I found that darlingtons are not the issue, the signal circuit is dominant over the bias point, it changes about 500mv over a range of 60°. The darlinton temp was set to 100° constant at 3.2A bias current (class A). Maybe it be worth to add a DC servo (chip opamp OPA27)

I changed de buffer with BD139/140 to increase the bias current of the buffer, which enhances the perfomance at high frec. ie:
THD @ 400KHZ full power lowers to 0.017, @ 100KC lowers to 0.034, @ 10KHZ 0.00078. Square wave is very clean with a slew rate of 17V/us.
I did some test using HEXFET's irf530/irf9450 (no lateral mosfets in Chile) but these devices are pure crap for quality audio at high frec., so I decided to keep the darlingtons. The dumping factor is 3800. I will desingn the servo and the PCB. I am an old BJT man and I feel BJT's perform a warmer sound than TUBES (mi old hamman kardon with KT-88 pentodes)

[CBS240]

Every transistor requires current to turn on. Once it is on, there is a finite amount of charge in the base of the transistor and when the input current is removed, there is a finite amount of time it takes for this charge to deplete. Or, you can apply a negative current to actively remove this charge to reduce the finite time it takes for the transistor to turn off by itself. At high frequencies this is required in order to better control the transistor, reduce distortion and prevent crossconduction.

If you physically connect the outputs to the BD139/140 devices (the ones with the collector connected to the rails) the heat from the outputs would cause their Vbe to decrease, thus providing a thermal feedback loop to the outputs. The problem is this may not be enough because the temp. coefficient of the Darlingtons is the product of the Tc from both devices inside the Darlington package. I once built a test circuit using Darlington OPT but I used a Darlington device for the thermal compensation. It still required fine tuning of the Vbe multiplier to accurately track the bias.

As for HEXFETs, if used as if they are conventional BJTs, they work as pure crap. BUT, Hexfets are not BJTs and work in a completely different manner. There are a few non-linear components, one being CISS and CRSS that corrupt the operation. These non-linear components are less pronounced in lateral fets but are still there. It is quite possible to build a very low distortion amplifier using hexfets or even planer stripe type diffusion mosfets that will work at 400KHz, but it requires a different approach and is not so simple to do correctly. They are designed as switching devices so there are inherent issues to deal with if one chooses to use them in a linear application. Many folks say they can't be used in low distortion linear amplifiers, but they are wrong. It just takes a bit more complexity, that which they are unwilling to do. For those folks, I suggest sticking to those nice linear expensive BJT's.

Incidently, if your understanding is up to it, here is an interesting paper regarding Darlington output transistors and how to improve thier operation.

[mnemneth]

Hi Artu,
interesting design but if I can be picky and if your software will allow, would you show your plots with a white background, please?

My tired old eyes find difficulty seeing the plot traces against the black background and if your amp is as good as it seems, I'd like to get as much information as possible.

Thank you,
Sandy