My first amp design - comments

Status
Not open for further replies.
First off all, I must say that English is not my native language, so you must excuse me for any grammar and spelling errors. 🙂

In past several months I have spent a lot of time in designing an amplifier for home use. I was inspired by a book named “Audio Power Amplifier Design Handbook” written by Douglas Self. Some parts of that book can be found on his site. After many hours... no months, of circuit simulation using Multisim 7 and freeware SwCAD III and later Multisim 10, I have come up with a nice, simple circuit, with more than enough power for home use. The circuit is presented at the bottom. It would be nice if someone could give me some guidelines or advices to improve the circuit. If you think that there are (or could be) some errors present in the circuit please give me a hint. In a month or two I would begun with PCB design.

Below is given simulated performace:

THD (1W, 8 ohm) 1kHz: 0,001% 20kHz: 0,002%
THD (16W, 8 ohm) 1kHz 0,001% 20kHz: 0,004%
THD (60W, 8 ohm) 1kHz 0,001% 20kHz: 0,004%
THD (95W, 8 ohm) 1kHz: 0,008 20kHz: 0,059%

Transistor Q15 is mounted on Q14 as a sensor for bias voltage. Small signal transistors used are 2SC2240 and 2SA970. Output devices are ONsemi products: MJL21193/21194.

If you have any question I would be happy to answer them.

Thanks for your comments and advices in advance.
 

Attachments

  • bipolar_v4.jpg
    bipolar_v4.jpg
    82.7 KB · Views: 1,614
Quickly viewed it looks good to me. I would suggest that you run the simulations using a circuit model of a realistic supply (one rectified from AC and having some series resistance at the power source). This often reveals some weaknesses in the design. I would also “isolate” the supply rails of the driver and VAS stages from the main supply rails with a RC filter. Likely a 100R and 100 uF would be all right.

I don’t know if there’s much help from the current limiter Q7 unless you current limit the output stage as well. In fact, at those power levels I suggest that you add this feature too. Also, connect two “anti-parallel” diodes from the base of Q1 to base of Q2. This will tie the inputs to about the same voltage potential even during clipping or slewing and thus protect the input from reverse biasing.

Just my :2c:
 
Hi,

Few comments:

- R21, R22, R28, R29 seems way too low for me. I'd use 0.33r there, then bias for 25...26mV Vre would be lower too, 80mA instead the 210mA (35W idle dissipation!). With that higher calue thermal stability would be better.
- R17 is too low, use R14 devided by three. Then change R17 to 22r.
- isn't R5, R6's value too ow? I'd switch to 47r or 68r (100...150mV voltage drop on them).
- for R3 and R4 emmiter degenerating resistors I prefer R=26/I(mA) > 12r.
- voltage reference for CCSs (formed by Q6) MUST be on the LTP's side!
- MJE340/350 aren't the best devices for VAS. Use 2SA1360 or similar (high Ft, excellent linearity, low Co) if you can get them.
- I'd change R16 to 1k for 1mA collector current for Q8.
- experiment with square wave oscillator and with an o'scope for the best value for C7, try 100n...1u
- you'd need a "fine tune" for C15, 10pF...100pF. Same for dominant capacitor ("C")!
- about R24: most designers prefer 10r, but I think 2.2r or 4.7r is enough there. For L1 1uH is enough.
- Teemuk, no need for isolation. VAS's and LTP's current topology gives excellent PSRR.
- what about adding a simpe OP short-circuit protection?
- rail fuses? 5A fast blow w'd do the job.
 
Hi Donk, I agree with all your suggestions, but I think Teemuk is right. I always use separate power supplies for the input and vas stages (one for the 2 (or 7...) input and vas stages, and one for each final stage). Infinite PSRR is the best ! Perfect ground wiring is an absolute necessity...

PL.
 
I made this amp strictly by the book, this is almost a copy of example amp given in the book. I have omitted protection circuit (simple single slope or multislope VI limiter) from schematic for the sake of simplicity. Because of that Q7 was added in schematic.

Teemuk, I will add anti-parallel diodes to Q1 and Q2. Good thinking! Thanks 😉

Donk:

R21, R22, R28, R29 were increased to 0,33 (I too tought that this was too low, but the book sad otherwise and Self in one of its amplifiers (class B) uses only 0,22ohm)
R17 changed to 22r (26mA current to switch Q7 protection, isn't this still to high?)
yet again, the book proposed 40-60mV as enough voltage drop on R5, R6 – changed to 68r (136mV)
I knew the R=(25 or 26)/I[mA] equation, it is even mentioned in his book, but he praised high value degeneration resistor as a way to linearise LTP. I was very doubtful if this is true, as no one (not majority) is using this high values.
Q6 in on LTP's side now
I can't find any spice models for 2SA1360/2SC3423, so I used MJE models. Didn't know of them until I saw PGP amplifier project found on this forum; will try to find them 😉
R16 changed to 1k
C7: need more time to get the right value

Thank you Donk! It looks like that with these new values the amp will deliver less THD at the same output power level. 1kHz went to 0,000% (can't increase precision to 4 decimals). At 95W and 20kHz I can sometimes see some artifacts in output signal (pic below) and THD rises, so C, C15 and C7 needs some more tweaking.

About separating power supplies for LTP and VAS from the output stage; I also think that there is no need to do such a thing, remember, this will be a home amp and not a high end design. Maybe I will add RC filter only in negative power supply line to deal with the problem.

Once again, thank you everyone! You were very helpfull 🙂
 

Attachments

  • output.jpg
    output.jpg
    90.2 KB · Views: 1,059
I can recommend KSA1381/KSC3503 instead of MJE340/350. These are excellent transistors from Fairchild that are ideal for VAS use. They also have SPICE models.

Instead of 2SA970/2SC2240 there is also 2SA1015/2SC1815, and Fairchild second source these as well (as KSA/KSC). They have good SPICE models for their version too 🙂
 
So, here is a new "improved" schematic not blindly based on Self's book

THD at 20kHz has dropped from 0,059% to about 0,02%, THD at 1kHz is barely measurable and sometimes shows 0,000% 🙂

Now I need to be sure that amp is absolutely stable and doesn't become an oscilator 🙁
so the capacitors C, C9 and C15 need to be fine tuned :scratch:
 

Attachments

  • bipolar_mod.jpg
    bipolar_mod.jpg
    88.5 KB · Views: 1,140
For me it looks better now!

As Cdom first I'd try 100pF, and then down to 82...68...47pF see what happens related to stability of the whole amplifier. Maybe you should try two-pole compensation, formed by 100pF and 470pF and an 1k resistor.
As C7 I'd try 220nF foil-cap, then watch 10...20kHz square-wave and sine-wave response of the amp, first at low (100mW, 1W) and then at high (100w) amplitudes.
C15 maybe sholuld eliminated. I wouldn't use it if the amp is absolutely stable with that. It can cause oscillations too, see the member called MJL21193's "Patchwork" thread.

Maybe you should try Microcap as simulator, I don't trust Multisim (I've got MS8 and MS10) for amp design.

PS.: if you can get 2SC4793 and it's complementary PNP pair then use them, they're better then MJE15030/31, and aren't so expensive devices. I found that MJE's beta difference is way too large between NPN and PNP.
 
I think your amplifier solves a problem you possibly didn't realise it could have . The VAS is a Darlington which nicely sidesteps a problem of the 0.7 V base voltage of the VAS transistor being very close to the typical 0.3 V of the current mirror ( saturated collectors plus R5 R6 volt drop ) . Although this is a transconductance stage it needs a little bit of room to swing .

When building a design similar to this I couldn't get the longtail pair current mirror to balance no matter what I did . Finally I used a SM component with very good results ( BCV 61 ? all in one current mirror/ emitter resistors as Douglas Self ) . The simplest solution was to use an emitter resistor on the VAS . Interestingly when using the BCV 61 the sound with or without the VAS emitter resistor was different . ( Using 47R at 7 mA VAS current ). The sound was punchier when the resistor was shorted ( 0R ) . I think all designers should pay attention to this, a big difference where none was expected . As the BCV 61 allowed it to work I settled on that . I wish this component was available in TO92 with 4 wires .

I think my observation completely endorses the views of Mr Self . Pump as much current as you can into the VAS I suspect the use of the current mirror is a very big deal in the amp . Again as it can't swing much in voltage terms , make sure it pumps plenty of current .

If you want to experiment, convert the VAS to a single transistor . You may find like I did that the current mirror is not so well balanced . Listen to the result ,especially without an emitter resistor . You maybe surprised . I thought I would hear nothing !

Short out R3 and R4 . Any reduction of hiss is sometimes better than a slight linearity gain of the input stage . Also linearity won at the input is lost in the output . Personally I want as much loop gain as I can get .

DC offset . Some think a little is a very good thing . If I keep it to about 20 mV I am happy with that .

Final question ., Why not use complimentary feedback pair as in the book ? Sorry if any of this is repetition of previous stuff, I only had a brief look ( should be working on a problem for my company ) .
 
A few more words to add . My design was an update of the old Hitachi MOSFET circuit . The longtailpair of 2SA 970 with ZTX 450 as current mirror using 22 R emitter swamping resistors ( BCV 61 better choice ) . The constant current source of the tail was a FET set at 2.7 mA . No additional emitter resistors between 2SA 970 ( re, internal 18R5 at this current )

The VAS 2SC2910 with FET constant current source set at 7 mA . No emitter resistor . More importantly a split cdom 100 pF ,220 pF and 2K2 as in the Douglas Self example ( page 192 ISBN 0-7506-8072-5 ) . For a MOSFET output ( Exicon TO3 10 N 10 P 160 V audio FET ) this is a powerful way to mop up the higher order distortion products .

When you are happy with your design I would recommend you try the two pole cdom on the VAS as in the book . I had no problems with it .

Douglas Self states and rightly so that MOSFET's are not the device of first choice . Having said that they do not require a Vbe multiplier bias , a simple fixed resistor will do it . Usually I set the standing current to 50 mA . I have even used my finger on the TO3 can as a good starting point ( 60 c ) . No source resistors are required ( i.e. emitter resistor for the outputs )

This very simple amplifier would put many high-end amps to shame . It can be built with nothing more than a volt meter . It is also very robust . A failed FET never seems to take out the driver stage .

I'm sorry I don't have a drawing . I think the information here is enough especially if following the book . The design costs about the same in parts as the Hitachi . It has no mystery feedback paths ( 22K and 500 R used as gain setting ,220 uF non polar capacitor ) Input side 22 K on the non inverting input with 2 x 10 uF ( 20 uF ) poly capacitor input . It out performs the Hitachi whilst retaining the special open sound quality . It is less fizzy sounding , I know the two pole cdom responsible for that . Also the current mirror adds punch , a quality MOSFET amps often lack .
 
PS.: if you can get 2SC4793 and it's complementary PNP pair then use them, they're better then MJE15030/31, and aren't so expensive devices. I found that MJE's beta difference is way too large between NPN and PNP.

Can I simply replace MJE's with 2SC4793/2SA1837 without changing the circuit? Would I need to change Cdom value also? The same question for MJE340/350 and 2SA1360/2SC3423 (or 2SA1381/2SC3503 as jaycee suggested)?

“Maybe you should try Microcap as simulator, I don't trust Multisim (I've got MS8 and MS10) for amp design. “

I only have only Multisim10 and SwCAD, Microcap is out of reach for me 🙁

“...see the member called MJL21193's "Patchwork" thread. “

I'm currently reading page 10 of the thread, very usefull 🙂


Instead of 2SA970/2SC2240 there is also 2SA1015/2SC1815...
I don't know which one to chose now.
 
Short out R3 and R4 . Any reduction of hiss is sometimes better than a slight linearity gain of the input stage . Also linearity won at the input is lost in the output . Personally I want as much loop gain as I can get.

I think it's good thing to make compromise between linearity and open loop gain, so I wouldn't change R3 and R4. Anyone? More comments on this?

Final question ., Why not use complimentary feedback pair as in the book ?

I love EFII topology, it's simple and performs nicely. 😉
 
I have used THESE as differential and/or mirror circuit without any emitter degeneration at all. A couple of drawbacks are that they aren't terribly cheap and I measured Hfe at about 90. However, they are fast, linear, and have a low noise figure. I find they work great at around 1mA. I'm sure there are other high quality matched arrays out there, but this one I use successfully. Cascode will solve the low Vce problem.🙂
 
I have had exams at my university so I didn't had much time for this. Well, I have done some minor changes in schematic and:
1. I included two pole compensation
2. LTP and VAS each has its own current source
3. I am considering dual devices like 2sc3381 and 2sa1349 for LTP and current mirror
4. output devices are changed to MJL3281/1302; 21193/4 are more suitable for subwoofer amp 🙂
5. an RC filter was added in negative power supply line to improve PSRR, there was little gain to include it in positive power line also.

I'm currently working on prototype PCB 🙂
 
CBS240 said:
I have used THESE as differential and/or mirror circuit without any emitter degeneration at all. A couple of drawbacks are that they aren't terribly cheap and I measured Hfe at about 90. However, they are fast, linear, and have a low noise figure. I find they work great at around 1mA. I'm sure there are other high quality matched arrays out there, but this one I use successfully. Cascode will solve the low Vce problem.🙂

I'm very doubtful I can find those devices here. 2sa/2sc3381 costs about 2 euros each here, and well known MAT devices are about 40 euro each :hot:
 
Status
Not open for further replies.