PWM -> Audio

You just lowpass the pin out of the fpga, usually just with one r-c. We mostly use this for things like trimming offsets or vcxo's, where the clock rate is huge compared to the filter cutoff, like 40 MHz clocks and 100 Hz filters. But there are lots of audio "1=bit" dacs, and I seem to recall they work at 2.8 MHz or something like that.

Lately people are going to multibit delta-sigma for better performance, but that needs a few-bit dac. This sort of blurs with the trick of dithering a dac to more bits.

D-S has the advantage that you don't have the pwm fundamental frequency screaming at you, and you can get very high resolution without being trapped by the clock/pwm ratio. The cost is a slightly elevated noise floor and a bit of distortion, and some more logic.

John

Yes, the DSM is free from the some of the PWM problems, however the reasonable quality audio requires the DSM clock to be at several MHz at the least. This is quite demanding for computation, as you mentioned. Also, driving the power output stage in the switching mode directly with DSM is problematic: high clock rate -> low efficiency and a hell of EMI. As for CVSD, that approach simply sucks. Combining the disadvantages of all modulations to produce a result which is poor by design.

Vladimir Vassilevsky DSP and Mixed Signal Design Consultant

I'll take your word on the math. The situation in the PWM case I was thinking of was using an 8051 to codewise do the PWM. Basically it was like this:

TimerInt: PUSH PSW ; Interrupt context entry XOR P1,#80H ; Reverse the bit MOV PSW,#TimePSW ; Gets my own reg bank MOV TL1,R1 PUSH ACC CLR A MOVC A,@A+DPTR ; DPTR is mine mine all mine MOV R1,A POP ACC POP PSW RETI

The table of values where the high and low duration alternating. It looked darn good on the scope after the filtering.

I just had a thought I'll fling out there to see if it bounces when it hits the ground.

Many bucker power supplies now use multiphase PWM. If multiple outputs of a FPGA updated each in turn, the ripple to be filtered out would be mostly at high frequency and yet each switching section would only be acting at 1MHz or so. It would also reduce the size of the devices in each section makiing high frequency operation easier to do.

Hey, I invented that about 1981!

John

The simplest way is through a resistor into a capacitor. Its then just a simple low pass filter. I have used to that good cheap effect !

If you invented everything, then who did invented you? Was it Jim Thompson or did you invented yourself?

VLV

Grammar, grammar!

My mom and dad invented me, of course, all properly married. Where did you come from?

I invented the dual-slope integrating A/D, too, when I was in high school, roughly 1963.

John

MooseFET snipped-for-privacy@rahul.net posted to sci.electronics.design:

Sure but OP said he was pincount bound already.

Vladimir Vassilevsky antispam snipped-for-privacy@hotmail.com posted to sci.electronics.design:

I fail to understand your dislike of CVSD. The telcos have been using since the 1960's and the telephones have remained perfectly intelligible. It did not even noticeably interfere with recognizing familiar voices. The bit rate is not all that high either (56k). Cell phone transcoders use only 14.4k and sound decent. Finally playing back transcoded waveforms is computationally cheap.

At one time, I made a mistake of using the CVSD codec. At 64k, it delivered the SINAD of only 20dB or so, with the ugly jitter on the transitions. The ordinary u-Law PCM codec performance is a lot better, and you can actually hear that. This behavior is a property of CVSD by design.

CVSD at 14.4k sounds terrible although intelligible.

Why using the modulation which is known to perform worse then any other modulation for the same data rate?

Vladimir Vassilevsky DSP and Mixed Signal Design Consultant

Only because he used the wrong micro :)

Vladimir Vassilevsky antispam snipped-for-privacy@hotmail.com posted to sci.electronics.design:

The literature that i have read does not support your claim that it is the worst at any given bit rate.

Vladimir Vassilevsky antispam snipped-for-privacy@hotmail.com posted to sci.electronics.design:

i have already asked you for references. Here in one from me:

The telcos may have used the CVSD at some long time ago however they quit using it since then. Why? Because it is not good.

I can see manipulations and speculations in this article. They are twiddling with the numbers and adding the empirical coefficients instead of doing the objective measurement.

Look at the datasheets on the CVSD codecs at CML web site (perhaps the CML is the only company which still does the CVSD products). Then compare it to the mere PCM codecs.

Vladimir Vassilevsky DSP and Mixed Signal Design Consultant

The army field phones use 16kbit CVSD.

Yes, they have used CVSD at the times of Alexander the Great. Julius Caesar adopted MELPe at 1200bit/s.

VLV

=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D I dont get it. You think I was joking? I was backing up your claim that CVSD was a valid compression scheme since someone else said it was a bad idea. Hopefully, use of hundreds of units in battlefiled conditions for years and years would give some validity to this claim. These are the phones that get thru on wires at the front. A truck with a shelter on the back is the PBX. "Mobile Subscriber Equipment" is what its called. Anyone else confirm?

BobG snipped-for-privacy@aol.com posted to sci.electronics.design:

You have it sideways Bob. Vassily says it is no good, though both u-law and A-law telephone companders are based on it. I say it is ok, but not great.