THAT transistor headphone amp (250ma discrete opamp) design sanity check.

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Hello Folks,

I was looking at my Mouser catalog the other day and noticed that they carried THAT corp matched transistors. I figured these would be a great input diff-pair + mirror and current mirrors for a driver stage.

The design goal is a headphone amp that can drive loads down to say 32-ohms (the average headphone is around 90-ohms). I figure a maximum of 250ma is very reasonable for the purpose.

I want the board to utilize SMT component as much as possible. I also don't want to bias the devices very high (no external heatsink) so I am thinking around 6-8ma. Using 3 pair of output devices allows the amp to still operate in class A for most of the volume range at which you would listen to headphones, yet still very good THD even when it goes into A/B.

Bias is adjusted with R22.

I chose the FZT1051A/FZT1151A from zetex for output devices because of the high current gain and low saturation voltage and good power capability for the task.

The circuit is designed for +/- 15V rails, but that is actually somewhat flexible.

I tried to isolate the power supply for the input/VAS/driver sections from the power transistors with R20/R21. This was to reduce supply ripple at those critical points.

R1/R2 would actually be a multi turn trimpot(20 ohm) to null the output offset.

Simulation gave me different results with the "gear" integration vs the "tap" integration. With gear the miller comp cap (C9) could be as low as 33pf and the amp would stay stable (in simulation) in trap it had to be at least 75pf. I will probably actually use 100pf for some safety margin.

It simulates very very well, but I just wanted to be sure it looks sane. Before I go buy parts and test it.

I have attached a PDF of the circuit.

Thanks for taking a look!

Cheers!
Russ
 

Attachments

  • ha1.pdf
    16.9 KB · Views: 677
Sim results

With 1V input 4vpp output 20khz, 32-ohm load:

Harmonic Frequency Fourier Normalized Phase Normalized
Number [Hz] Component Component [degree] Phase [deg]
1 2.000e+04 2.000e+00 1.000e+00 179.88° 0.00°
2 4.000e+04 7.479e-07 3.739e-07 -58.92° -238.80°
3 6.000e+04 5.004e-07 2.502e-07 -101.83° -281.72°
4 8.000e+04 4.467e-08 2.233e-08 32.37° -147.52°
5 1.000e+05 1.239e-07 6.197e-08 -89.18° -269.06°
6 1.200e+05 2.436e-09 1.218e-09 -135.74° -315.63°
7 1.400e+05 1.010e-08 5.050e-09 -100.76° -280.65°
8 1.600e+05 4.799e-10 2.399e-10 -150.25° -330.13°
9 1.800e+05 3.181e-09 1.590e-09 3.94° -175.95°
Total Harmonic Distortion: 0.000045%
 
Hi Russ

I have used this very part, http://www.thatcorp.com/300desc.html 340 SOIC 14, and they work very well. The only thing is the Hfe, which I measured at about 88. You might consider a CFP or constant power arrangement for the input differential. These transitors don't like lots of current. My best results were with Ic between 0.8mA to 2mA. Changing the simple current mirrors to Wilson mirrors would keep a constant voltage on both THAT transistors. If you do this, go for low noise devices. I like http://www.fairchildsemi.com/ds/KS/KST5088.pdf and http://www.fairchildsemi.com/ds/KS/KST5087.pdf They have high gain and bandwidth, are cheap and also available from Mouser. They could be used as Q's 5, 16, and 17 as well. I measure Hfe for these at close to 400 with a DMM.:)
Another advantage is that I was able to use the THAT arrays without any degeneration. The price you pay for with Hfe, I guess, but the essence of a current mirror is not gain, it is equality.
 
Hi Russ,

looks good, I'd say, epescially the output diamond buffer. Maybe you want to consider HF degeneration of the LTP (emitter Rs, shunted at the emitters with a coil for AF) to get an additional pole/zero in the open-loop gain. I also second CBS's comment on the Wilson mirrors but then you will need to lift the emitter of Q5 a bit (that is, lower it actually, e.g. with a bypassed resistor or a diode drop).

- Klaus
 
CSB240, and KSTR thanks very much for your input.


I have been experimenting in SPICE with your suggestions.

I also decided that probably one of those THAT300 devices per channel is OK, the mirrors for the output buffer can be degenerated sufficiently the matching should not be too crucial. I used 10R which should be enough don't you think?

Here is the result. :)

4vpp 20khz 32ohm load (about 62ma peaks) output devices biased at 7ma.

Harmonic Frequency Fourier Normalized Phase Normalized
Number [Hz] Component Component [degree] Phase [deg]
1 2.000e+04 2.000e+00 1.000e+00 179.82° 0.00°
2 4.000e+04 5.317e-07 2.659e-07 81.75° -98.06°
3 6.000e+04 1.589e-07 7.947e-08 -122.71° -302.52°
4 8.000e+04 4.046e-08 2.023e-08 50.80° -129.02°
5 1.000e+05 5.966e-08 2.983e-08 -77.56° -257.37°
6 1.200e+05 3.747e-09 1.873e-09 8.69° -171.13°
7 1.400e+05 9.068e-09 4.534e-09 -71.58° -251.39°
8 1.600e+05 3.267e-09 1.634e-09 95.26° -84.55°
9 1.800e+05 4.653e-10 2.327e-10 125.60° -54.22°
Total Harmonic Distortion: 0.000028%

Here is the new circuit.
 

Attachments

  • aha1.pdf
    36.7 KB · Views: 205
you might want to use a realistic input, load impedance, especially for estimating stability

I've played the earlier circuit and found next to no phase margin when I added 1 nF to the output to sim cable load C, and upped the Vin source to a few KOhm (being lazy I only used the "simple" Middlebrook test not the Tain loop gain probe - the loop gain probe may give different results)

I'd add 100 Ohm series with 100 pF from in+ to gnd and change the sim Vsource to include the reasonable midpoint resistance of the Volume pot

since it appears you're willing to play with the mirror/VAS, I think the ef buffer between them is a good deal - huge open loop gain boost - can be arranged for good symmetry, allows reasonable values of mirror Q degen to reduce their noise contribution

C10,11 just seem like a bad idea, C2 is too big, the closed loop gain should probably be higher, smaller area, faster, lower C Qs in the driver add a little to the phase margin in my sim
C1 has a hard time helping when LED incremental resistance is likely 10s of Ohms, its better for ps rejection to split R14 and bypass the midpoint, if you want to filter diode noise and isolate the 2 sides RC filters from the diode to Q bases are better

with these fast Q there seems to be no problem using Cherry's output stage enclosing compensation, I always like to try 2 pole to max audio band loop gain

V4 is the added "Middlebrook" test AC source, plot V(out)/V(fb)
or go for the full loop_gain_probe in the examples/Educational/LoopGain file
 

Attachments

  • jcx_ha1.asc.txt
    10.7 KB · Views: 117
oops, that wasn't so good - I know better than to largely design with the .AC loop gain in Spice, it can really give misleading results

I think this patches up most issues - I cut one output pair since I would use the 2.5 W rated sot223 pkg
 

Attachments

  • jcx_try2_ha1.asc.txt
    12.6 KB · Views: 84
Wow, a lot to chew on JCX. :)

Very good info, thanks for the interest and help.

I am still thinking I would like to drop the second THAT device (at the diamond buffer) in favor of degenerated driver Qs.

Let me see if I can guess the purpose of Rs 5,6,23,24,31 and C15,C16.

I am guess that Rs form a voltage divider to help control the bias.

And that the C's help keep the impedance from the diamond buffer front end low at audio Fs. Correct?


Cheers!
Russ
 
Extreme redesign. :)

Will be knocking out a prototype tonight.

I like this design better because it seems to have much better behaved offset. R12 and R13 are a multi turn pot with the wiper to VEE.

I have not decided yet to use JFETs CCSs for simple resistors to set the mirror currents. Using JFETs would allow more predictable circuit performance with varying supply voltages.

Also for now I will not use the THAT devices simply because I have none on hand for tonight's build. :)

Anyway here is the schematic and the LTSpice simulation should anyone want to play with it.

Cheers!
Russ
 

Attachments

  • ha.pdf
    33.4 KB · Views: 205
More time to sim... And a working prototype....

I increased the size of C4 to 75pf and added 100ohms in series to give better phase margin. I also added the same to the input to act as a low pass.

I am still new to SPICE so I hope I used the probes correctly.

Now I get the following results:

Open loop gain: 70.8db

Phase margin at unity gain with 2.2K feedback R and 6pf comp is right around 36 degrees. The AOL falls of at about 28db per decade as it crosses 0db.

At 6db gain phase margin over 45 degrees and the slope is right about 20db.

Open loop gain reaches 0db at 10Mhz.


So I am guessing in this configuration (with the change in C4) the amp should be stable to unity and very stable above 6db.

Last night rigged a prototype on veroboard, and it worked just fine.

I have a couple of boxes of 2N5550 and 2N5401 I bought very cheap, they actually worked fine, though I am sure there are better Qs for the job.

I matched my Qs for gain, but not for anything else. Offset was <2 mv with the adjustment pot centered. It only took about half a turn to get it to zero.

Now I will refine it a bit and put together a proper headphone amp.

Cheers!
Russ
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.