Attenuator.

[Diabolical Artificer]

In order to connect the OP from a power amplifier to a spectrum analyser or soundcard/DI I need to attenuate the signal but without loading or changing the load of the amp. Also don't want the attenuator to change the signal in any way. One last thing it has to deal with a IP of between 1v to 70v RMS ( maybe not that much but it would be nice). We also want to be able to vary the OP from 0 - 1v RMS tops.

The easiest way is to use a resistive divider, but having mucked about with this approach a bit, the results arn't brilliant. I used a 10M/1M divider, at first with a 5M pot in parallel with the 1M R. I changed the 1M to a 4M7 in parallel with 5M pot, which is better. This is ok ish for a 10v RMS IP but crap for 1v. If I reduce the big resistors value leaving the pot and 4M7 as is, this is ok for a small IP, but this reduces the "load" as seen by the amp.

So what I'm after is a constant IP Z, so thought about an opamp like the TL072, with a FET IP this has a big IP Z, but you can't stick 70v on it's IP. What about a tfmr? I have quite a good selection with a few that are good from 50hz to 20khz, so could try a resistor in series with the pri to knock the current down, and then use a potential divider on the secondary to vary the OP, but this will change the Z on the primary.

How would you go about it?

TFL, Andy.

[Mark Hennessy]

There are two separate things here: the dummy load, and the divider. Don't confuse them.

The problem you're having is because of the resistor values. Why use such large values? Think about stray capacitance, cable capacitance, and input capacitance (of the measurement gear) - all will play havoc. Also, they will have a lot of Johnson (thermal) noise.

Next, the input impedance of your measuring device will load down the output. If it's got an input impedance of 1M, it might not be significant. But 10k - typical for a soundcard - will. In effect, the input impedance is in parallel with the lower resistance value in your potential divider.

I'd build a simple 100:1 attenuator based on the lowest value resistors I can get away with. The lower you use, the less "loading effect" you'll get.

Here, "loading effect" means the change in output voltage with the input impedance of the measuring device.

Perhaps make the lower resistor 100 ohms. Then the upper resistor would need to be 9.9k. Obviously, you'd have to put some in series and/or parallel to get to this value - you might even include a pre-set to calibrate it.

The downside of low value resistors is that they will dissipate some heat. Overall, those resistances will dissipate about 1 watt with a 100V RMS input. But given that you're using a number of them to get to 9.9k (say, a pair of 4k7 and a 500 ohm - a 1k pre-set resistor would be my choice there), then standard 0.5W resistors will survive. Upgrade to bigger ones if you can. Prefer series combinations rather than parallel, as resistors have a thing called voltage coefficient.

The output impedance of such a network is essentially 100 ohms (actually 99), so a change in Zin of your measuring equipment from 1M to 10k will make about 0.08dB of difference.

In terms of high frequency performance, because the divider has such a low source impedance, stray C won't have much effect. Even if you had 100pF at the output (cable plus Cin of the measuring device), the -3dB point would be 16MHz.

Needless to say, putting 10k in parallel with your existing dummy loads will have no significant effect on the power amplifier. Don't worry about it.

Finally, if you want to make it variable from 0 to 1V, replace the bottom 100 ohm resistor with a 100 ohm potentiometer. Simple! Personally, I like calibration, so I'd retain the 1k pre-set, and calibrate it so that it was definitely 100:1 when at the top end of the track.

[Diabolical Artificer]

Thanks Mark. Think we may have got our wires crossed? Anyhoo, didn't think about the IP Z of any test gear. I was more concerned about the attenuator which sits in parallel with the dummy load, changing the value of the load.

In the end I built a 100:1/ 10:1 stereo switchable attenuator using a 68k and 22k both 2% to give 90k and a 22k preset to get 99k and a 1k 10 turn pot. I used what I had, I had no 100 ohm pots. A SPDT SW switches between 10 and 100:1.

"I'd build a simple 100:1 attenuator based on the lowest value resistors I can get away with. The lower you use, the less "loading effect" you'll get." I'm confused about this, surely a higher R or Z "load" is better, hence why scope and DMM have 1M or 10M Z. So if our test gear has a 10k Z and we use a 100 ohm R in parallel then any signal will reduce by 100 times? If we have 10k the signal will be halved, no? I'll admitt I'm struggling to visualise this.

As regards HF losses though a lower Z is better. With 10p of capacitance in a cable say inc stray C, there'd be 3dB loss at 159khz with a 100 ohm Z in parallel, I hope I have that right. But as were measuring audio frequencies, who cares.

BTW, I measured the DC res of my soundcard and it's 10k, so the Z will be a bit less than this. As it is the attenuator works well so far. I been mucking about with various PC based SA's as I mentioned. Rtspect though very simple shows harmonics. Anaprof though more sophisticated is only a demo version and I think has certain features locked. You can't change the IP range, I couldn't get the bugger to work properly and it turns itself off after 6 mins. The SA on audacity is good, though you can't see a signal "live". You have to record it. REW is quite nifty, you can do frequency response graph's and it "nulls" out the frequency glitches in your soundcard/DI. Not sure if it has a THD measurement doodad in it, but heard the sig gen is good. Need to play more.

Andy.

[Mark Hennessy]

Thanks Mark. Think we may have got our wires crossed?  Anyhoo, didn't think about the IP Z of any test gear. I was more concerned about the attenuator which sits in parallel with the dummy load, changing the value of the load.

Hi Andy,

No crossed wires - I understood that you were primarily concerned about loading down the output of the amplifier (see my penultimate paragraph), and that is why you initially chose such high value resistors, but I was trying to explain that that wasn't the issue - rather, the loading effect at the output of the attenuator would be the problem

In the end I built a 100:1/ 10:1 stereo switchable attenuator using a 68k and 22k both 2% to give 90k and a 22k preset to get 99k and a 1k 10 turn pot. I used what I had, I had no 100 ohm pots. A SPDT SW switches between 10 and 100:1.

These values sound quite reasonable. My suggestion was only an example to give us some numbers to discuss.

"I'd build a simple 100:1 attenuator based on the lowest value resistors I can get away with. The lower you use, the less "loading effect" you'll get." I'm confused about this, surely a higher R or Z "load" is better, hence why scope and DMM have 1M or 10M Z. So if our test gear has a 10k Z and we use a 100 ohm R in parallel then any signal will reduce by 100 times? If we have 10k the signal will be halved, no? I'll admitt I'm struggling to visualise this.

I've knocked up a diagram that should explain this. In essence, there are two interfaces where Loading Effect can take place, and we have to consider them separately (remember that I was focussing on the interface between the output of the attenuator and the input of the test gear).

Basically, it shows that we can model a real amplifier output stage using a "perfect" voltage source (with zero output impedance) and a series resistance. And an input stage can be modelled as a "perfect" amplifier (with infinite input impedance) and a parallel resistance. Simplifications, but very useful.

   

As regards HF losses though a lower Z is better. With 10p of capacitance in a cable say inc stray C, there'd be 3dB loss at 159khz with a 100 ohm Z in parallel, I hope I have that right. But as were measuring audio frequencies, who cares.

Yes indeed. That's the point I was making - keep the impedances low, and stray C is not an issue. But it would have been with your MΩ resistances.

As a conservative rule of thumb, figure on about 100pF per metre for coax cable. It might be less in practice - easy enough to measure it needed.

BTW, I measured the DC res of my soundcard and it's 10k, so the Z will be a bit less than this

Not necessarily. But it might be.

Measuring AC impedance of a simple thing like this is quite easy - would you like me to talk you through it? Hint: you only need a resistor (or variable resistor) and some basic Ohm's law...



In summary, while we've discussed loading effect before, it's worth reiterating that whenever you connect anything to anything, there is some loading effect because the device generating the signal will have an output impedance, and the device receiving the signal will have an input impedance.

This change in signal level varies from negligible to disastrous, and includes values that are necessary and expected (such as 6dB, the loss you get when you have to match impedances).

Usually, when making measurements you want the loading effect to be as small as possible, as you don't want the measurement itself to change the conditions you're trying to observe. I know that much is stating the obvious, of course - we all know that a DMM has a high input impedance (10M or more), and that an AVO could cause problems in some situations.

Continually assessing the possible impact of loading effect is something that is just second-nature to experienced engineers. But I know from my day job that many graduate engineers don't arrive with this ability - it's something that takes a bit of time to cement. So don't feel bad if you get confused! Tomorrow in class I'll be explaining all of this from first principles (and there will be homework and a test!), and I know that while some will get it immediately, others will struggle to understand and apply it.

I hope all this helps,

Mark