At least a few times I wished I had a "soundcard on steroids", so that I can replicate a specific waveform I'm getting from a generator, transformer or other devices I'm dealing with (e.g., see this question).

I read this is called an arbitrary waveform generator.

The twist is that I need a high-voltage one and high-power one, reproducing amplitudes to about 250V. On the other hand I don't need high frequency (or frequency precision/stability), so a soundcard followed by ×200 voltage amplification will do the job.

Example specifications:

  • Frequency: up to 1 kHz, arbitrary waveform, may have DC offset;
  • Voltage output: ±250V;
  • Current capability: 3A or 300W, whichever is lower (e.g. 100V@3A or 250V@1A);
  • Output impedance: < 5Ω .

How can I make such a device? I'm all for buying off-the-shelf parts to reduce work. E.g. the DDS could really be a soundcard (or a computer-controlled DAC).

For the power amplifier I thought about using a big audio amp, but I'm worried about two problems: they are generally lower output voltage (designed to drive 2Ω or 4Ω loads), and probably cannot tolerate DC offset (I think most, if not all, audio amps use capacitive coupling at the input).

I also found the LME49830 which can be used to build the device I want (albeit "only" ±200V output). It too worries me that it is shown as capacitively-coupled.

What would you recommend?

  • \$\begingroup\$ Buffer + big heatsinks. \$\endgroup\$
    – Long Pham
    Jun 10, 2018 at 15:43
  • \$\begingroup\$ Your sound card can't generate DC, so AC coupling to the amplifier isn't a problem. \$\endgroup\$
    – JRE
    Jun 10, 2018 at 15:45
  • \$\begingroup\$ If the soundcard doesn't cut it, I'd use a computer-controlled DAC. The thing is that I want to be able to generate waveforms that have DC bias. \$\endgroup\$
    – anrieff
    Jun 10, 2018 at 15:52
  • \$\begingroup\$ Ignoring the DC requirement, how low does your AC signal need to go in frequency? \$\endgroup\$
    – Andy aka
    Jun 10, 2018 at 17:21
  • \$\begingroup\$ Have you looked at e.g. Apex PA93? As an alternative to the LME4980. Expensive, but if this is a one-off... \$\endgroup\$
    – Jack B
    Jun 10, 2018 at 18:12

5 Answers 5


I'm going to suggest the same PWM scheme as Marcus, in the off-the-shelf version, ie, a Class D audio amp.

You can find IRS2092S based class-D amps on aliexpress at very cheap prices. This is a driver chip which allows +/- 100V power rails, and it uses separate MOSFETs, so pay attention to the voltage and power rating of the amp you're buying, as this will depend on the which parts are on the board... you will also need a power supply, most likely a switching one.

Since the voltage you need is pretty high, I'd use a stereo amp in bridged mode, which would give +/- 200V, followed by a transformer.

These amps can be DC coupled without trouble, however this is of course not compatible with a voltage-boosting output transformer.

If you want a DC-capable amp with more voltage, then something like IRS2092S followed by a MOSFET driver of proper voltage specification sounds like a nice starting point.

Note that this will give you a voltage. If you want to emulate an alternator or something like that, you'll also have to emulate its output impedance.

If class D is too noisy for you, then you're going to need a very heavy chunk of metal, as an amp capable to output 250V even with a lowish current of 1A quickly results in tons of output transistors for reasons of SOA (Safe Operating Area) especially if the load is inductive. In this case I wouldnt bother, just get a used PA amp from the pawn shop and remove the input DC coupling cap.

Another option I'd consider is a Class-D PA amp with an optical (or at least transformer isolated) SPDIF digital audio input. Maybe Behringer iNuke has that, I don't know. Having the digital to analog conversion inside the amp, and having it isolated from the PC with optical fiber, could end up being a good idea if the high voltage experiment encounters some "unforeseen consequences".


Puh, top of my head: isolation transformer 1:2 (or 1:N, really) with center tap to convert a single-phase household grid voltage to something with a definite "0" as well as actual +- 230 V after a bridge rectifier; followed by said bridge rectifier.

Then, add a halfbridge, i.e. one IGBT / MOSFET from the +230V rail to output, and one from output to -230 V. (you could have rolled with a full H-bridge instead, and worked with a single-sided supply – but that has implications on your ability to ground your device under test).

Connect that output to a load resistor (say, some 3Ω?), followed by a capacitor bank against reference. Make sure things can't get too hot. If you overdimension your switches' protective diodes, you might be able to use inductivities instead of resistors – but getting a 1000V 3A inductivity might be a challenge of its own.

Address the switches in your bridge with a PWM; your load resistor/capacitor bank act as smoothing low pass filter. The higher the PWM frequency, the smaller-valued the resistor and capacitors need to be to achieve a given maximum ripple – but the worse the EMI you have to fight, and the more losses in your switches you'll have to cool away.

Control said PWM with a feedback loop, observing the output voltage, and comparing it to the "should be" instantaneous voltage of your desired signal. The PWM frequency will inherently set an upper limit to your signal frequency, but in the single-digit kHz range, you're probably fine, theoretically.

Benefit of all that is that the real part of your output impedance is really only limited by your transformer's ability to transport power, and your switches' maximum current, as you regulate the voltage of the whole system at its output. Also, unlike a linear regulator, you don't "burn" all the voltage you don't use.
Downside is, quite obviously complexity (control loop fiddling... yay!) and cost (transformer's not free, nor are high-power high-voltage switches nor capacitors).

  • \$\begingroup\$ PWM at 500 V must be a real something, even at 10 kHz (to satisfy 1 kHz design requirement). Think of switching times and transients to switch between 500-V levels. Sorry, :-( \$\endgroup\$ Jun 10, 2018 at 18:05
  • \$\begingroup\$ @AliChen I was afraid that would end up being a real design constraint, but then I drafted that system: If I restrict myself to N-channel MOSFETs with a breakdown voltage of >=600 V, and a continuous drain current > 10 A, I do get a lot of potential switches. So, I restricted myself to things with Rdson <= 100mΩ, I get Si-based MOSFETs with less than 20 nC of gate charge. Afterwards, I wasn't really afraid to say you can switch these with upwards of a dozen kHz. But I might really be missing something here – never build such a thing. \$\endgroup\$ Jun 10, 2018 at 18:17
  • \$\begingroup\$ @AliChen so, what's the limiting factor here? \$\endgroup\$ Jun 10, 2018 at 18:17
  • 1
    \$\begingroup\$ Technically the ordinary computer power supplies do the switching at 340-V level, at 50-100 kHz I believe. But they do have a sophisticated control over the switching trajectory in I-V space. So the D-class at +-250V is likely doable, but needs some deep engineering experience and real care. I couldn't find any industrial DC-1kHz-band amplifier in 250-V range, but I wasn't looking for more than 30 minutes. I believe the main reason they don't exists because there is no need for such high voltage, and the power can be achieved by having loads with less impedance. \$\endgroup\$ Jun 10, 2018 at 19:02
  • \$\begingroup\$ I think you'd probably encounter similar dU/dt speeds in ultrasonic welder supplies – these drive massive piezos at high speeds. \$\endgroup\$ Jun 10, 2018 at 20:49

The outlined "problem" has a two-step solution:

Step_1: get a AWG - arbitrary waveform generator, typically based on computer-controlled DAC, with standard, say +-5V, output.

Step_2: connect it to a power amplifier. For +- 250-V output you will need to design your own using big (and possibly cascoded) transistors at the output stage of an ordinary operational amplifier, just as LME49830 suggests, only with bigger transistors and some additional level translation.

Alternatively you can modify-reuse a stock audio amplifier, like this one from Markertek if you find a 250-V output. This will be a challenge, even for a 100-V output.

The AC coupling can be ignored and worked around - all power stages in audio are DC amplifiers, but with artificially-cut bandwidth at DC end, to prevent damages to acoustic drivers. You will need to tap into DC input, and maybe design some front-end bias circuit.

For the bandwidth of 1 kHz there might be "programmable power supplies" with sub-millisecond time response. Again, finding a 250-V one would be challenging and expensive, for a reason.

Yet another alternative is to use amplifiers used to drive shakers/exciters in vibration testing. This Bruel&Kjaer model 2721 can drive DC to 15 kHz with output up to ~100 V (71V RMS). Higher voltages must be a challenge, since the loads tend to align with reasonable transistor voltages, and not to arbitrary +-250V.


You may want to look into topologies and components used in VFDs. Essentially, a VFD is just a computer-controlled waveform generator followed by a class-D amplifier operating at hundreds of volts and tens of amps (small ones, larger ones involve larger voltages and currents) followed by an output filter. The device typically produces waveforms at frequencies up to 1khz. This is a hot topic nowadays and there is a ton of information on the net. For a design example you can refer to this application note from Microchip, but there are other sources too.

For the purpose described you'll likely want to omit the motor control math and concentrate just on the amplifier and output filter. It is usually realized with a three phase bridge driver IC driving some MOSFETs or IGBTs. There are H-bridges too.

All that remains is to generate 4 or 6 PWM signals from your microcontroller to approximate desired waveform.

With regard to powering the circuit... I believe a big transformer is unavoidable for safety reasons.


Despite the price point the Keysight B2900 series SMU would do exactly what you need



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.