Class D Switching Amplifier

[dmuxaf]

I'm familiar with the basic theory of operation for class d switching amp's. But, for my project, I am inputting a modulated signal with a 5kHz bandwidth centered at 25kHz. I'm still waiting on some specifics from the customer as to what the signal is and type of modulation. Anway, I have tested the amp with a pure 25kHz sine wave and everything works as calculated (~8% difference), which leads me to believe the amp is fine. But, I'm nowhere close with the modulated signal input. First, I have about a 50% difference in calculated power (25kHz) versus input power across the load (input producing more power). Lowering the on-delay and switching frequency produces even more power across the load, but I still have a 50% difference from calculations. I know it's a modulated input signal, but 50%?!?!? Maybe that's normal, but I can't find any reference material to help back this up. Also, I can't find any resource for calculating/estimating switching frequency. I had it at 300kHz at first so I know I could block it with a filter and still have fairly linear gain across the BW.

Anyway, any insight/thoughts is appreciated.

 

[geekEE]

Check to see if your power supply rails are sagging when you put in the modulated signal. Is the output going to heat up and increase resistance to any appreciable amount? Are you output chokes too high in resistance compared to your load? Is your modulation index too low to get your desired output?

Glenn

 

[dmuxaf]

I should have mentioned this, sorry. But, the load is 500 ohms with a .01uF in series (odd, I know) and I need ~50W at the load.

I checked the power supply rails and they are fine (i.e. no noise, sagging, etc...). I don't believe the output should or does heat-up and change any resistances. I'm also using 100W resistors. The inductors forming the output filter measure at ~.5 ohms. And, my modulation index is 71%, which I could probably tweak to around 80-85%. I'm also using a full-bridge setup.

Again, with a pure sine wave input, everything works fine. What considerations should I take into account when using a modulated input signal? I didn't think it would be that much different, and maybe it shouldn't.

Thanks for the help.

 

[geekEE]

That's really an odd load. It's a high-pass filter with a knee at 32kHz. Your 25kHz sine wave is already below the knee and I assume the modulating frequency is going to be much lower than that, which means that very little will get through. Sounds like you'll need really high voltages to pass power through this thing! And the distortion will be really high.

This is not a load that I'm used to driving. I wonder if you could design your output filter so that combining it with your .01uF cap created a bandpass filter at your modulating frequency. Of course, this again would be more power at the cost of more distortion. What's the goal? What are your parameters?

Glenn

 

[dmuxaf]

I work, intern, in an R&D lab and was given a signal (.wav format) from a potential customer. They want the signal they provided to be passed through our amp onto the load they requested at ~50W. No, I did not run the .wav file out the sound-card; I'm using a NI-DAQ.

I calculated and simulated the filter stage, including the load, and had resonance ~70kHz and cutoff frequency of ~125kHz - everything below that passing. I looked at this across the load in the simulation. I attached a pic of the simulation from LTSpice.

They said the signal was already band-passed; however, I looked at the FFT of the input waveform using matlab and labview and saw a whole bunch of higher frequency crap, so I put a bandpass filter (20kHz-30kHz) at the input of the of the comparator to knock anything out.

I can get the signal amplified to 35-40W with a much lower switching frequency and on-delay than expected (150kHz and 150ns respectively) with a bus voltage of 350-400V. BUT, the voltage and current at the load is much higher (~50%)than calculated values, which I calculated using 25kHz.

This may all be fine and operating correctly with a modulated input; I just like being able to understand/explain things, and something tells me this shouldn't be this way.

 

[geekEE]

Does your bandpass filter attenuate the signal any within the range of interest? How are you measuring the power on the output and comparing it to your bandpassed input signal? i.e. is the stuff that you're filtering out part of the reduction in power? Do you know if that is an accurate model for the load? Did you take the modulation index into account in your SPICE simulation?

Glenn

 

[VEBill]

If I'm reading this correctly, the confusion is surrounding the discrepant power measurements. And the two signals being compared are the pure 25kHz 'carrier', versus the same carrier when it is being 'amplitude modulated' with a 5kHz bandwidth signal.

This sounds like the same situation when dealing with AM transmitters (and thus my AM 'keywords' above).

When an AM transmitter with a carrier power of 100 watts is transmitting but unmodulated, the output power is 100 watts. When it is 100% modulated, the peak output power reaches 400 watts (putting a great strain on the system), and the average power will be about 200 watts. The waveform shape can affect the average power.

In other words, measuring the power level of non-trivial waveforms can be non-trivial. And this can even apply to the signal generator used to create the input signal.

So, if it isn't the PS rails drooping, then it might be this issue.

 

[dmuxaf]

I used the LTSpice simult simply for frequency response. I initially used it when putting together the output filter. I wanted to block the swx freq and have somewhat linear gain over the BW of concern (22.5kHz-27.5kHz), which, according to the LTSpice simult, the range is between 7.3db-7.8db.

After this, I made some calculations based on using an unmodulated 25kHz sine wave, since this was the center frequency of the BW. I have no idea what type of modulation this signal employs, and I'm now waiting on additional info from the customer. So, when using a 25kHz sine wave from a func gen and comparing to my calculations (load power and bus voltage), I was about ~4% off (sorry, not ~8%). I simply have a diff probe on the load in addition to a current probe. From these measurements and the PF, I'm calculating Pavg.

I then moved to the .wav file (centered at 25kHz with a 5kHz BW) provided and sent it out through a NI-DAQ. I tried it using different swx freqs and different on-delays. The lower each of these are, the more power I get across the load.

What's confusing me is, with the same swx freq and on-delay, the 25kHz sine wave from the func gen generates more power at the load than the modulated signal does. I actually thought the modulated signal would generate more power, regardless the type of modulation. But, it's less.

 

[VEBill]

The AM 'discrepancy' (not quite the right word) would perhaps explain more, but not less. So it's not that.

 

[geekEE]

What are the diff probe and current probe connected to? Are they true-RMS meters?

Are you using the customer's actual device, or are you substituting a resistor and capacitor for their device? The reason that I ask is that if you are using their actual device, it may have a more complex model than the simple representation that they gave you. If that's the case, try using a resistor and capacitor and see if you get different results.

Glenn

 

[dmuxaf]

The diff probe is connected across the load, which consist of actual resistors and a cap; the customer did not supply anything other than the .wav file. The current probe is attached to the line through the load (i.e. output of amp to load). I am using an oscope to take RMS measurements of Vload and Aload and then taking the mean, via the oscope, for Pavg.

 

[geekEE]

Your scope calculates true RMS for non-sinusoidal waveforms? That's pretty nice.

I'm starting to run out of ideas here. Here's a couple of things. First of all, try measuring the power through the 500-ohm resistor alone. That should take any power factor considerations out of the equation. This should be about the same as the power you measured across the entire load.

Secondly, neglecting power altogether, how close is your measured voltage gain to your calculated voltage gain for the system?

Glenn