PWM to PWM Converter . . . But No Microprocessor

Is this for a speed sensitive steering assist setup? or what?

Good guess! But no. See my post #6.

Thank you, Rixter. I too was thinking of a simple RC filter and driving pin 5 of a 555. I may breadboard that and see if it is good enough for my purpose. It's a touchy-feely application; I'll know if it's OK or not pretty quick.

I am trying to understand what you are trying to do but it is not clear at all.

I guess the 10-100Hz signal is your control signal. You probably mean that this is your bandwidth, but why PWM? The only explanation I can think of is that you are using digital communication in order to achieve some noise immunity.

Anyway, with that signal you want to drive your servo at 400Hz PWM. Here you will need to average from cycle to cycle. So your bandwidth is reduced to 5Hz.

Your response time will be 200ms. If you want to improve that, you must have forgotten the k in your input signal, 10kHz to 100kHz.

Anyway, servo motors do not require such a fast response, so I would be surprised that you need that. Increase the speed of your input signal (it does not make any sense to be this slow) and stick with the 400Hz modulation and it should be fine.

I guess carrier frequency is the same as periodicity? So it would be 10hz-100hz, constanting changing (it's an EFI pulse from an ICE running at 600-6000 RPM). Can't put a number to the output resolution yet, but guessing within 1% should be OK. And output needs to be real steady as it is driving a critical servomotor in real time.

Wish I could reveal more, but I have a partner with patent fever. Dream on...

I'm game to try a micro. The 8051..I'll go read up on it and see if I have the right stuff for the job

Thanks for all the good pics and info. What do you think of Arduinos and such?

Yeah, I grew up on analog and feel comfy with 555s and op amps. Here's a nifty chip that looks like it'll do the voltage to PWM: v to pwm

Gee . . . Linear took all the fun out of it.

Nice chip. I like Linear's offerings. I've used a lot of their stuff all over the place.

Think of each cycle of you EFI PWM as a sample of the input. By Nyquist, the highest frequency which you can produce is 1/2 the sample rate. So your EFI signal will be less than 5 Hz at 600 RPM and less than 50 Hz at 6000 RPM. I'm sure in reality it will be a lot less than this.

If you filter the PWM signal to convert to a slowly varying EFI voltage signal, at slow speed, you could get some ripple which you do not want to pass on to your 400 Hz PWM output. Another problem is a low pass filter will result in some latency.

Here is a circuit from an Electronic Design News article that converts the PWM to an analog voltage without the ripple or latency. You can then drive your voltage to PWM converter with the output.

http://www.edn.com/design/analog/4347834/Circuit-converts-pulse-width-to-voltage

(There are ICs that do this job, but the frequency specs only go down to 30 Hz. You might want to try one to see what it does at 10 Hz.)

http://cds.linear.com/docs/en/product-selector-card/2PB_2645f.pdf

This looks really good, but I see that it gives a voltage proportional to pulse width, not duty cycle. Not sure that will work for us...gotta noodle on it. Maybe some mod to that circuit would do it. Thanks for the idea!

I have one of those PWM-to-voltage chips that is spec'd down to 50 hz and I'll see what it does at 10hz.

You are correct, the EDN circuit needs a constant rate, not variable like yours.

Here is a wild idea that just might work. Your ideal pwm-to-voltage converter would produce a voltage after each cycle, regardless of how long the cycle is (variable frequency). The voltage value is proportional to the duty cycle, the ratio of the pulse time to cycle time.

1. Convert the pulse time and cycle time to voltages using integrators, integrating a constant value for these times. (Switch on rising-edge, falling-edge, next rising-edge)

2. Feed pulse width and cycle time voltages through two log amps

3. Get difference of log cycle time and log pulse width with difference op-amplifier.

4. Run this difference through exponential (antilog) amplifier.

Vout = (pulse time) / (cycle time)

Vout = exp ( log(pulse time) - log(cycle time) )

https://ocw.mit.edu/courses/electrical-engineering-and-computer-science/6-071j-introduction-to-electronics-signals-and-measurement-spring-2006/lecture-notes/23_op_amps2.pdf

Better yet, here's an IC that will do the voltage ratio for you:

http://www.analog.com/media/en/technical-documentation/data-sheets/AD534.pdf

Rixter, you and Brave Sir Robin have given me much to think about. I need to review my system and do some testing and see how to proceed. Your wild idea is an ingenious design and I am most thankful. CR4 works great when good engineers gather with thoughtful inputs.

I'll post some results and thoughts soon. In the meantime anyone is most welcome to chime in.

Just adding top Rixter's post #3

The output frequency is 400Hz, and, you want a response of 1ms?

Not nearly possible.

You won't get 1ms response from a conversion to a 400Hz pwm from a 10Hz - 100Hz PWM in this reality. I believe Nyquist has something to say about this.

Never mind!

I've been busy. I built a PWM generator (PWM generator) with variable freq and variable duty cycle so I can work on the bench. But the idea of just using a LP filter to get a voltage proportional to duty cycle won't work; the duty cycles I'm looking at go down to 2% and no amount of filtering can turn that into a stable voltage. But I've dreamed up a circuit that might work:

When the PWM is high it drives a switch that charges C1 thru R1. When the PWM pulse goes low another switch connects C1 to C2 momentarily. Making C1>>C2 ensures that the voltage on C2 matches C1. While the PWM is low C1 discharges thru R2 and waits for the next pulse, and the cycle repeats. Now V is not a linear function of duty cycle, but that doesn't matter in my application so long as it is a monotonically increasing function.

See any problems with this or improvements? Now I just need a V to PWM to drive the servo at a fixed frequency. I think the LTC6992 will work.

I'd get rid of the switches and replace switch two etc. with a standard peak detector

Forget it. I messed up; this circuit won't do the job. Back to the drawing board.