Microcontroller Class D Amplifier, Rev2

11 March 2012

A problem of the first revision of this amplifier was the limited precision at low volumes. Although the PWM of the ATtiny45 is just 8-bit, I realized that it can produce arbitrary precision output. Timer1 can be clocked at 64 MHz from the high frequency PLL, which combined with the double-buffered counter compare value and the overflow interrupt, can get 9 and even 10-bit precision in the audible range. Lets say we want to DAC the value 300/512. We set OCR1A to 255 (which will hold the output high) and on the timer overflow, we write 300-255 = 45. Writing this 9-bit value takes 2 timer cycles.

I made a few other small improvements:

• Replaced the BJTs with MOSFETs. This reduces ADC noise by removing the loads at the output pins of the MCU and also drops the component count because we no longer need the base resistors.
• Added a filtering capacitor at the ADC ARef port (which is now available after replacing timer0 with timer1). This allows us to use the internal 2.54V ARef voltage for 2.54Vpp input.
• Added a better output LC filter to protect the speakers from switching noise.

I experimented with 2 output stage driving strategies:

• 0% duty cycle at idle. This is the most efficient driving method because only one side of the H-bridge is switching per cycle. It didn't work too well in practice because at low volumes we might get just 1 pulse in 512. The raise and fall times of the drivers were not fast enough for such a short pulse introducing error (this pulse has too high frequency harmonics for my implementation).
• 50% duty cycle at idle. It might seem as a waste, but note that at idle both sides raise and fall simultaneously, so no current goes through the speaker. A small downside is a little power loss at the cross-over Vgs of the power MOSFETs. This strategy sounds perceivably better, though.

Dithering is no longer required, but still the output interrupt got a bit longer.

// 2x 9-bit PWM. 50% duty cycle at idle.
u8  time;
u16 pos;
u16 neg;

ISR(TIMER1_OVF_vect) {
    if(1 & ++time) { // every odd cycle
        u16 adc  = ADC;
        s16 val1 = (adc >> 1) - 256;
        u8  val2 = (u8)(adc & 1);

        pos = 256 + val1 + val2;
        neg = 256 - val1;

        if(val1 >= 254) // prevent wrap-around
            pos = 510;

        if(val1 <= -255)
            neg = 510;

        if(pos > 255) {
            DACP = 255;
            pos -= 255;
        } else {
            DACP = pos;
            pos = 0;
        }

        if(neg > 255) {
            DACN = 255;
            neg -= 255;
        } else {
            DACN = neg;
            neg = 0;
        }
    } else { // every even cycle
        DACP = pos;
        DACN = neg;
    }
}

The LEDs are all functionally required - I use them as zener diodes.

The following links contain a picture of the schematic and the eagle schematic + C source code.

Circuit diagram.

See the first revision for a video.

Comments for Microcontroller Class D Amplifier, Rev2

Danny Flam on December 29, 2012

Then again, http://www.wpi.edu/Pubs/E-project/Available/E-project-042711-190851/unrestricted/PWM_Techniques_final.pdf Interesting idea: drive each half of the output devices separately, using 3 level PWM and you will have even less noise. might be a simple modification? but then you can do 3 level PDM and get even higher accuracy... :)

Alexander on December 30, 2012

Hi! could you just send me the hex file on e-mail please? I can't seem to compile the source code available on the webpage. Thank you!

Matic on January 29, 2013

hello. I was wondering and it is not mentioned if this is a mono or stereo amplifier. It seems like it is mono or monoblock? Could it be run at higher than 70w power? Thank you for the answers.

Bob C. on January 30, 2013

I just got done breadboarding this. The sound is not bad. I had to substitute IRFZ34 and IR9Z34. Couldn't find a reasonable price on the others. They run hot using a 12vdc supply. I assume this is because the RDS is lower than the IRFZ14. Adjusting the 12v supply down to 10v allows it to run cool with no heat sinks. The only major snag was programming the fuse bits on the attiny85. You have to make sure to switch the clock to PLL. Cost me a little time because I'm not an Atmel expert. Having fun with it. Thank you

Shaun on February 11, 2013

How can I get this to work on an Atmega32? Several functions come up as undeclared when I compile.

Frederik Crevits on February 21, 2013

Why do you use LED's in the circuit?

zigulinis7 on February 22, 2013

What should be the fuse bits? Should I check PLL?

Bob C. on February 23, 2013

These are the fuse bits I used: http://www.frank-zhao.com/fusecalc/fusecalc.php?chip=attiny85&LOW=E1&HIGH=DF&EXTENDED=FF&LOCKBIT=FF

Tom on March 7, 2013

There are 4 more z-diodes in the picture.I can't finde them in the schematics? Can you update your schematics.

brian on March 29, 2013

can you show me how the 4 diode is connected?

Kstutis on April 5, 2013

Hello :) can i use any of power fets instead irf9z14 and irfz14? can i use any fet for bs170? what is THD for this amp??? thanq you. p.s. higher Vdd higher the power? if i use 50V instead 12v?

Rouslan on May 28, 2013

Tom & Kstutis, I used LEDs instead of zener diodes just for fun (and to see what's going on). To increase the supply voltage and get higher power, you'll need to pick appropriate zener diodes to bias the gates of the power transistors such that they never turn on simultaneously. I don't recommend that however, since the ADC is not precise enough for higher gains. It will be nice exercise to get negative feedback and actually ADC the difference between output and input. This might be hard with the bridged circuit, but should be easy with a single ended one. Finally, you could use any transitors as long as you bias them correctly. This doesn't need to be too precise and likely the circuit will work as is. If the power FETs get warm, use zeners with slightly higher breakdown voltage.

Post a new comment