PWM and analog ouput

Terry Given ha scritto:

I guess I can't use the LMV824 in this particular case because I need an output span of 10V (-5V / +5V).

I used several times the mfb filters with no problem. I'm going to try also this way, adding a -vref to bring down the output and selecting the appropriate gain to reach both ends.

Thanks to you too Marco / iw2nzm

I first noticed this with exactly this type of circuit. I was only aiming for 8 bits, but used an LM324 and could clearly see switching edges on the output - about 100mV in a +/-10V output. not huge, but enough to warrant fixing. two 0603s and voila, gone. and a 3rd-order response, as well.

A bit later on I was working with a 3rd-party controller, which had the Sallen-Key circuit but used a 12MHz GBW amplifier. And the feedthru was still there, but a few mV.

sure. but by choosing multiple feedback over SK, the problem goes away.

what the 2nd order SK filter requires when driven with a PWM waveform is that the opamp GBW >> Fedge (where Fedge is the nominal edge frequency, a few MHz), rather than simply GBW >> Fpwm (in this case a few kHz).

which is another way of saying: look at the complete spectrum to be filtered.

I havent done so, but you could probably arrange an SK design so C2 >> C1 and R2

100n Maybe ... || ,-----------||-----------, | || | | | | +5V | ___ ___ | ___ |\\| | 3V3 PWM i/p o---|___|--o---|___|--o---|___|---o------|+\\ | DCout 100k | 100k 100k | | >--o-----o | | |-|-/ | +/-5V | | | |/| | 330n| 150n| | -5V .-. --- --- | | | --- --- | | |20k | | | '-' -o- -o- | | 0V 0V --------o | .-. | |10k | | '-' | -o- +2.5V

(created by AACircuit v1.28 beta 10/06/04

It's 3rd order with a 10Hz break. With only +/5V supplies you'd need a rail to rail opamp. Higher supply voltages and use pretty much anything. Uses 2 more components than a 2nd order version but myself I've found 2nd order filters a waste of space for this type of job.

the accuracy comes from the stability of the OSC of the uC. You should have a offset pot or a way to trim the ref for all + inputs of the OP-amps for example.. I used a single ref that was trim able for all 4 op-amps in a 324 type package. dithering the scaled variables will give you resolution increase over what you may lose with using a single hardware timer and scaled variables. I used A WORD (Double Byte) field , the low order was the fraction while the upper the whole. the idea is that at some point, you'll have one less or one more variable count before it hits it's mark to change the output.. if you are using a rather high speed carrier frequency, doing this every 4 or 5 cycles by varying the width will give you an over all average change. You can actually simulate a very high resolution DAC converter as long as the response time does not have to be that fast.

--
http://webpages.charter.net/jamie_5"

My extensive experience in active filters for telephone use in the early '70's says the above configuration will NOT behave as expected... wrong shape and corner frequency, due to dissipation factor of the capacitors.

See...

for an example.

(Paste URL into browser, the tail end doesn't look like it might address automatically.)

...Jim Thompson

--
|  James E.Thompson, P.E.                           |    mens     |
|  Analog Innovations, Inc.                         |     et      |
|  Analog/Mixed-Signal ASIC's and Discrete Systems  |    manus    |
|  Phoenix, Arizona            Voice:(480)460-2350  |             |
|  E-mail Address at Website     Fax:(480)460-2142  |  Brass Rat  |
|       http://www.analog-innovations.com           |    1962     |
             
         America: Land of the Free, Because of the Brave

If you do use a sallen-key, look at the open loop output impedance of the op-amp at high frequencies. Some op-amps have quite high output impedances at high frequencies others tend to stay low. In general rail-to-rail opamps have higher impedances beyond their gain bandwidth because the output is from the collectors of transistors.

If you use a "c-load" rated op-amp, you can put a capacitive load on the op-amp to help matters.

Jamie ha scritto:

I need to keep out any pot. It's by design. Anyway, the required accuracy is quite poor: < 1% over the full-range.

Ok, I got it.

Thanks Marco / iw2nzm

Thank you for articulating so clearly what i suspected was right.

Implicit in all the linear operational amplifier circuits is that the drive is within their range of response. Fast edge square wave drives are usually out. The OP can buy an relatively cheap octal "digital date filter" in ZIP, these are L-C-L tees and will take the edge off. Depending on his internal resolution, gain errors can be compensated for by scaling the duty cycle...

That's cheating-)

Which?

Pasting the URL, or padding out dissipation effects?

Dissipation effects at low frequencies are far more significant than one would think. Discovered it quite by accident while futzing over a

30Hz high pass that was acting weird... did some equation cranking, and... voila ;-)

...Jim Thompson

--
|  James E.Thompson, P.E.                           |    mens     |
|  Analog Innovations, Inc.                         |     et      |
|  Analog/Mixed-Signal ASIC's and Discrete Systems  |    manus    |
|  Phoenix, Arizona            Voice:(480)460-2350  |             |
|  E-mail Address at Website     Fax:(480)460-2142  |  Brass Rat  |
|       http://www.analog-innovations.com           |    1962     |
             
         America: Land of the Free, Because of the Brave

----- Original Message ----- From: "Jim Thompson" Newsgroups: sci.electronics.design Sent: Sunday, March 02, 2008 4:11 PM Subject: Re: PWM and analog ouput

lower

rail

2nd [...]

Nice idea. You must have had some tight telcom spec's to meet, (or p**** poor '70's caps :). Biggest annoyance I've had with cap' quality is when asking a bit too much from the numerous gyrator designs. Loss compensation can sometimes help but it always feels 'tricky dicky' and second best.

With the S+K above (working Q about 1), cap' Q's even in the 10 area will have little effect. I'd say what's more at stake here is the need for 'wholesale' as opposed to 'technical' filtering. The smaller the ripple the better. Although the S+K AC response is sensitive to component variations (infamous), the set-up is good for the PWM DC gain, has a fast(ish) rolloff and a small(ish) part count.

Apologies Jim. One day I might get this Outlook Express thing sorted out.

[snip]

Real men don't use Outhouse Excuse.

Try Eudora v7.1.0.9

...Jim Thompson

-- | James E.Thompson, P.E. | mens | | Analog Innovations, Inc. | et | | Analog/Mixed-Signal ASIC's and Discrete Systems | manus | | Phoenix, Arizona Voice:(480)460-2350 | | | E-mail Address at Website Fax:(480)460-2142 | Brass Rat | |

| 1962 | America: Land of the Free, Because of the Brave

^^^^^^^^

well now its explicit. IMHO its these sorts of "implicit" behaviours that cause problems. and its easier to learn from someone elses experience than in your own product.

sure (might wanna look at damping though). you could probably even buy something from minicircuits, or any of a variety of expensive things. hard to beat an 0603 R & C though.

Cheers Terry

absolutely. its just this sort of application (Fsw = 1-10kHz) can easily lead a user to erroneously assume a low GBW opamp (LM324) can do the job, when its GBW is 10-100x too small.

Terry Given ha scritto:

I admit I never thought this and I was going to use the LM324 :) Well, this thread is very interesting. I learned a lot.

Thanks Marco / iw2nzm

de

Probably not. Provided that clock period doesn't vary significantly from tick to tick - and they rarely do - this is mostly unlikely to be a significant elemet in your error budget. One of my acquaintances once did a 23-bit accurate D/A based on pulse width modulation, and his worst problem was charge soak in the integrating/filter capacitors, even using Teflon (PTFE) dielectric and minimising the charge being integrated.

f the

So why use a 324 - untrimmed ultra-cheap - opamp, where a trimmed part is available for less extra money than you'd pay for the pot, let alone the twiddling to get it right.

The amplitude of the dither component is pretty low to start with, and provided the frequency of the dither is still at least an order of magnitude higher than the 3dB frequency of the low pass filter on the output, you won't need to be obsessive about maximising the frequency of the dithering events.

-- Bill Sloman, Nijmegen

snipped-for-privacy@ieee.org ha scritto:

For example, what op-amp should I use? I need a dual-supply (typically

-12V/+12V) and it should be available to RS or Farnell.

There are so much ap-amps out there, one is disoriented when selecting a device.

Bye Marco / iw2nzm

I have the same problem. If you aren't using the parts frequently and keeping an eye on the latest offerings, you have to devote a day or two digging through the various manufacturer's lists.

to find low offset op amps, and then dig some more to see who stocks them and how much they are.

If you post the right query here you can often get a pretty good answer.

My favourite low offpset op amps are the Linear Technology LT1013 dual and the LT1014 quad, though they are big, old and expensive so it's probably not an answer that you should accept - the OP101 is probably cheaper and good enough. There will be even worse answers, along with better ones.

-- Bill Sloman, Nijmegen