Simple Class D Speaker Amp Using a MOSFET Driver

[Phil Pilgrim (PhiPi)]

There have been several threads regarding speaker output for the Propeller. Most involve analog amplification of filtered DUTY-mode output from one of the counters. There's another kind of amplifier called "Class D" that sends PWM output directly to the speaker, letting the voice coil's inductance and speaker cone's physical inertia do all the filtering. It occurred to me that a MOSFET driver chip might have enough oomph to drive an 8-ohm speaker. If so, it would then be a simple matter of driving its inputs with the proper signals to drive the speaker differentially, H-bridge fashion:

http://forums.parallax.com/attachment.php?attachmentid=62005

Here's the circuit I used. I selected a Micrel MIC4469 becasue it had some useful logic on its inputs. It also has built-in protection diodes on the outputs for inductive loads. Notice that two outputs from each side are paralleled to provide more current drive. The "sign" input basically selects which side of the speaker receives the DUTY modulation. When the sign is set (negative audio swing), the duty will be more than 50%. That's just right, because it goes into the inverting input of the lower half of the driver.

http://forums.parallax.com/attachment.php?attachmentid=62006

Here is the test program. It generates a sinewave, which my 6W 8-ohm speaker reproduced with very reasonable volume. It was loud enough to annoy my cat Browser from his top-shelf sleeping perch (photo below) and send him out the door. (He needs more exercise anyway.) The driver chip did not even get warm, which is one of the advantages of Class D amplification.




CON

_clkmode = xtal1 + pll16x
_xinfreq = 5_000_000

duty_pin = 0
sign_pin = 1

sine = $e000

VAR

long waveform[*64]
byte cogno

PUB demo | amplitude, i, phase, t, f

repeat i from 0 to 63
waveform[*i] := sin(i << 7) * 32767
amplitude~
start(duty_pin, sign_pin, @amplitude)
f := clkfreq / 30000
t := cnt
repeat
waitcnt(t += f)
amplitude := waveform[*i := (i + 1) & 63]

PUB start(dutypin, signpin, ampaddr)

'' Start the class D output driver.
'' dutypin: Prop pin number (0 - 27) for DUTY output.
'' singpin: Prop pin number (0 - 27) for SIGN output.
'' ampaddr: Hub address of a LONG containing the instantaneous audio amplitude.
stop
|< dutypin
|< signpin
return cogno := cognew(@classD, @dutypin) + 1

PUB stop

'' Stop the class D output driver.

if (cogno)
cogstop(cogno - 1)
cogno~

PRI sin(x) : value

'' Sine of the angle x: 0 to 360 degrees = $0000 to $2000

if (x & $fff == $800)
value := $1_0000
elseif (x & $800)
value := word[*sine][*-x & $7ff]
else
value := word[*sine][*x & $7ff]
if (x & $1000)
value := -value

DAT

org 0

classD mov amp,par 'Assumes A0 and A1 for output.
rdlong dutymask,amp 'Get the duty pin mask.
add amp,#4
rdlong signmask,amp 'Get the sign pin mask.
add amp,#4
rdlong ampladdr,amp 'Get the amplitude address.
mov dira,dutymask
or dira,signmask

:loop add phs,amp wc '[* 4] Add amplitude value to software counter phase.
muxc outa,dutymask '[* 4] Carry is duty output.

rdlong new_amp,ampladdr '[* 1 + 7] Get the next amplitude value.

add phs,amp wc '[* 4] Add current amplitude value to software counter phase.
muxc outa,dutymask '[* 4] Carry is duty output.

shl new_amp,#1 wc '[* 4] Multiply new amplitude value by 2, saving the sign in carry.
muxc out,signmask '[* 4] Set the sign value in out.

add phs,amp wc '[* 4] Add current amplitude value to software counter phase.
muxc outa,dutymask '[* 4] Carry is duty output.

add phs,new_amp wc '[* 4] Add new amplitude value to software counter phase.
muxc out,dutymask '[* 4] Carry is duty output: save in out.

nop '[* 4] Delay for constant 200ns pulse timing.
mov outa,out '[* 4] Output new duty and sign simultaneously.

mov amp,new_amp '[* 4] Now, set current amplitude to new amplitude.
jmp #:loop '[* 4] Ad infinitum.
'----
'[*64] Clocks for loop, or four hub cycles.

phs res 1 'Phase of the software counter.
amp res 1 'Cog copy of amplitude.
new_amp res 1 'Cog buffer for new amplitude.
out res 1 'Buffer for outa.
dutymask res 1 'Pin mask for duty output.
signmask res 1 'Pin mask for sign output.
ampladdr res 1 'Hub address of the instantanous audio amplitude.




I tried connecting the MIC4469's Vdd pin to Vin and, as expected, the volume got much louder, but with some audible distortion. I have a feeling that there are inductive issues which cause the distortion that need to be addressed somehow. (Perhaps Beau can weigh in on this issue.) Anyway, it was an interesting experiment which proves the principle, at least, that minimal-component Class D audio amps are possible, given the right kind of output from the Prop.

-Phil

Addendum: I changed the code to make it a proper object and doubled the DUTY output level, effectively qudrupling the available output power. It made a big difference, and I didn't hear the distortion I did with the higher (Vin) supply voltage. I also changed the schematic to include pulldowns on the driver inputs. This will keep the driver in a quiescent state when the Prop pins are tri-stated, protecting the driver from a continuous over-current condition. For maximum protection, capacitively coupling the Prop outputs might also be recommended.

Addendum 2: I changed the source code once more to update the DUTY mode ouput at regular 200ns intervals (assuming 80MHz clock).

_

Post Edited (Phil Pilgrim (PhiPi)) : 7/4/2009 8:51:30 PM GMT

[mctrivia]

Cool. Now can the wav player be converted to use this?

▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
propmod_us and propmod_1x1 are in stock. Only $30 (http://propmodule.com/?x=products). PCB available for $5

Want to make projects and have Gadget Gangster sell them for you? propmod-us_ps_sd (http://www.gadgetgangster.com/160) and propmod-1x1 (http://www.gadgetgangster.com/161) are now available for use in your Gadget Gangster Projects.

Need to upload large images or movies for use in the forum. you can do so at uploader.propmodule.com (http://uploader.propmodule.com) for free.

[Phil Pilgrim (PhiPi)]

It should be possible for any program that outputs audio to use it, although this test program is not a real object and needs massaging to make it one. Most audio-generating programs write to a counter's frqx register. This can be changed by doing a wrlong to the amplitude hub variable instead.

-Phil

[localroger]

Dude, the post office owns your cat?

[Phil Pilgrim (PhiPi)]

No, but my cat thinks he owns the Post Office; therefore the box is his. Who am I to argue with that?

-Phil

[kwinn]

That's a great idea Phil, and fyi that is just about how variable speed 3 phase motor drivers work. Of course you need 3 circuits for that.

[Phil Pilgrim (PhiPi)]

I've made some changes to the code, as noted in the first post, to increase the volume level.

-Phil

[Rayman]

You really should try to play music with it some how... I recently had a problematic audio circuit that would play sine waves just fine but played music horribly...

▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
My Prop Info&Apps: ·http://www.rayslogic.com/propeller/propeller.htm

[BTX]

@ Ray
Due the armonic components for a high quality audio signal, I think you should turn the PWM period so short as possible, considering the max MOSFET capacity, and the bigger frequency you would like to play. I don't know if the prop could do this.

>>> I think you should turn the PWM period so short as possible... (Or better... do some math before, to calculate the correct PWM period)

@ Phil
What is the sinewave frequency that your demo plays ? 1Khz ?

▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Regards.

Alberto.

[Phil Pilgrim (PhiPi)]

Alberto,

The sine wave frequency is more like 400Hz. That's about the fastest I could generate in Spin using 64 samples per cycle. I used a sine wave to test because it's easier to hear any distortion than with something more complex. The minimum duty-mode pulse width from the driver itself is about 675ns, which seems to get filtered out adequately.

-Phil

[Phil Pilgrim (PhiPi)]

I tried the driver with a speech demo and also at lower amplitudes. The lower-level sounds do introduce some distortion (i.e. "fuzz"), primarily due to the low (~1.5MHz) sample rate. Although 1.5MHz sounds high, consider that a single pulse every 256 samples represents a 5.8KHz signal, which is audible. Using a counter to generate the DUTY output might seem to be a solution, but there the pulses are shorter than the frequency response of the MIC4469, which has rise and fall times around 50ns. It's just not capable of responding to a 12.5 ns pulse, and there's no way to clock a DUTY-mode counter at less than the system frequency. It may be possible to strike a compromise using a video driver, but I've not tried that yet.

-Phil

Post Edited (Phil Pilgrim (PhiPi)) : 7/4/2009 8:59:22 PM GMT

[Chris Micro]

Hi Phil,

it seems to me that the mc4469 is not so common. Imho you could to the same with a 6 times Inverter.
On one side you can connect the PWM-signal and the other side is the polarity.
At zero crossing you would have to exchange the high/low pulse width durations.

chris

[Phil Pilgrim (PhiPi)]

Chris, it's the MIC4469 (Micrel, not Motorola), and DigiKey stocks them. A regular logic inverter will not work. It has neither the output capacity nor the built-in protections required for driving a speaker.

-Phil

[Chris Micro]

Hey, I didnt take a look at the datasheet of this Ic. 1.2 Amps is a lot of current. For a lot of applications your solution will be very good.

But I'm shure, there can be something done with less.
For instance one could use a 74hc244 and connect the buffers parallel. Probably 2 capacitors will be needed to decouple the speaker. The current can be up to 140 mA.

[mctrivia]

1.2 amps may sound like a lot but it is only 11.5w

▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
propmod_us and propmod_1x1 are in stock. Only $30 (http://propmodule.com/?x=products). PCB available for $5

Want to make projects and have Gadget Gangster sell them for you? propmod-us_ps_sd (http://www.gadgetgangster.com/160) and propmod-1x1 (http://www.gadgetgangster.com/161) are now available for use in your Gadget Gangster Projects.

Need to upload large images or movies for use in the forum. you can do so at uploader.propmodule.com (http://uploader.propmodule.com) for free.

Post Edited (mctrivia) : 7/4/2009 7:41:59 PM GMT

[Philldapill]

This is really an "out of the box" kind of idea. I like that.

I have quite a few 9A MOSFET drivers that work very well. They are UCC27322 from TI. If you need more current handling ability, give 'em a try.

EDIT: Also, Phil, I think those rise/fall times are a function of the load capacitance. The driver I mentioned above, has rise/fall times of about 20ns, but that's with a 10nF load. The rise/fall times you mentioned are for a 1nF load(datasheet). If you are only driving the speaker, and no capacitance, don't you think that these rise/fall times would probably be faster than listed in the datasheet? The propogation delay would still be the same, but that wouldn't matter too much if ALL drivers had a delay.

Post Edited (Philldapill) : 7/4/2009 8:11:19 PM GMT

[Phil Pilgrim (PhiPi)]

I've modified the source (main post) to update the output at regular 200ns intervals. This higher update rate has helped to reduce (but not entirely eliminate) some of the fuzz that was audible at lower amplitudes. A single pulse every 256 intervals (8-bit PWM-like) is equivalent to 19.5KHz, which is out of hearing range for most people.

-Phil

Post Edited (Phil Pilgrim (PhiPi)) : 7/4/2009 9:00:45 PM GMT