XR2206 revived by China ?

[ez24]

That would be intersting to compare but i do not have time for this .

Bummer

[setq]

Just a heads up with those Chinese XR2206s. Before I bought a proper one, I had an XR2206 based function generator that I built myself. The thing worked fine up to about 1MHz with no distortion other than what can be expected from a cheap diode clamp sine generator.

However when changing the output transistors after a little accident, I manage to blow up the XR2206 by shorting the output with a probe tip.

So off to eBay, bought another one and used direct substitution. Total shit. Distorted wave output above about 300KHz and there are switching transients at the peak and trough of the wave. Plus it got very hot and the integrator is impossible to control at the high end of the frequency ranges.

So I thought screw this and bought a discrete 1980s Exact branded unit from eBay and won't touch an XR2206 again.

I also won't buy any crap from China any more. RS for parts and eBay for branded test gear only.

[David Hess]

That would be intersting to compare but i do not have time for this . However from reading i infer it is easy to validate by cranking up the supply voltage. It seems Chinese clone gives up somewhere pass 12v mark.

That makes sense.  The Chinese copy is probably built on a modern low voltage analog process instead of the obsolete 36V NPN junction isolated process that the original likely used.

[stj]

wouldnt a modern MCU like an ARM4 or ARM7 with a fast dac be able to leave these things in the dust?

i'v seen diy generators based on Nucleo/Discovery boards that have pre-calculated wave-tables streaming to the dac under dma that look very fast.

[setq]

This is pretty much how AFGs work although there is usually an ASIC with a LUT or something in it instead. Think the AD9850 works the same as well. They are much more expensive than cheap analog function generators as they have some difficult problems to solve with harmonic content, clock division, control, feedback etc. Plus once you've solved all of these you still have a driver amp on the end of it, a power supply on the front end. Literally 50% of the device is still analogue.

Personally I find that full analogue generators have a nice manual control which fits better with some use cases than encoders or manual frequency programming. Sometimes you need to frequency sweep by hand. I have the RF equivalent of an AFG (Marconi 2019A, repaired yesterday whoohoo!) i.e. it does sine only and it's pretty much useless for sweeping frequencies. If you want a spot frequency and power level/amplitude you're fine. If you want to sweep by hand forget it. If you want to sweep automatically, AFGs have some nice features but you can do that with two analogue generators with the VCF in of the second one connected to the ramp out of the first one.

I'd still like a nice HP 33120a, but it's not happening.

[stj]

that marconi sounds like it was designed for jamming radio signals! 

[setq]

It probably could be set up to cause some trouble with much as anything RF related. The RF output power isn't particularly high though so you'd need a big PA and a noise source to modulate it against. By which time Ofcom and the fuzz would probably kick your door down and take it away so that sort of stuff is a self-deprecating game.

I'm using it for testing IF amplifiers and keeping the living room warm at the moment which is something it can just about handle.

[dannyf]

wouldnt a modern MCU like an ARM4 or ARM7 with a fast dac be able to leave these things in the dust?

depends on your set-up. I think 16-points per cycle is probably the minimum for most people. that means a 16Mhz F_CPU chip can output a waveform of 1Mhz waveform. Plus overhead and stuff like that, you are likely to be much slower.

[bktemp]

wouldnt a modern MCU like an ARM4 or ARM7 with a fast dac be able to leave these things in the dust?

i'v seen diy generators based on Nucleo/Discovery boards that have pre-calculated wave-tables streaming to the dac under dma that look very fast.

It depends on the waveform and how much "postprocessing" is done in the analogue domain: For a sinewave with a good (brick wall) filter, you need 2 samples per cycle and still get a good looking sinewave (the amplitude will drop a bit due to sinc rolloff). For any other waveform you need a much higher samplerate, because of the harmonic contents (Nyquist-Shannon).
A FPGA is probably much better suited than a microcontroller for DDS because DMA can't do variable steps between the waveform table. The only solution I can think of is calculating the waveform in software and using DMA only for transferring the samples from a FIFO to the output ports at a fixed rate.
Even a cheap FPGA can easily do >100MS/s for at least 20MHz sinewave and 1-10MHz squarewave, triangle or arbitrary waveform (depending on you signal quality requirements).
I wouldn't use a microcontroller for anything above 100kHz.

[stj]

the ARM thing i saw was a lot faster than that, it was playing a loop sample from ram via dma with no cpu overhead, the dac can exceed 80MHz afaik.
cpu was clocking at either 120 or 180 MHz.

[dannyf]

Try to load up a table to the DAC, and allow a user selected fine tuned output frequency  and then you may appreciate the challenges.

[Kleinstein]

DDS sine generation uses constant output rate, so DMA for buffering and software to calculate the values is ok. An FPGA is just faster than most µCs (single cycle compared to maybe 5 to 15 with an µC).

One can do DDS in software in an µC. There are some AVR based solutions that go up to around 2 MHz output sampling and thus a maybe useful maximum output frequency of up to 500 kHz (depends on the filter). However this is only 8 Bit data and one still needs the DAC. Also the µC is busy with output, so any change in frequency will cause glitches.

So it should be possible to use one of the ARM µCs with integrated DAC to make a reasonable low cost sine generator up to maybe 2 or 10 MHz, depending on the DAC. You still need the filter, attenuator and amplifier, just like with an fully analog FGen. The filter part sounds complicated, but is can be just a passive LC filter combination, if the clock is not too slow.

[bktemp]

the ARM thing i saw was a lot faster than that, it was playing a loop sample from ram via dma with no cpu overhead, the dac can exceed 80MHz afaik.
cpu was clocking at either 120 or 180 MHz.

If the samplerate is fixed, the available output frequencies are limited to samplerate/n where n is an integer from 2 (for sinewave with brick wall filter) to number of sample that fit into available memory. You can increase the resolution for higher frequencies by putting multiple cycles into the loop buffer, but the resolution is still limited to samplerate*m/n. And you need to recalcutate the sample memory whenever the frequency is changed. So you can't change the frequency without interrupting the waveform.
The big advantage of DDS is the almost infinite frequency resolution: You have a frequency resolution of samplerate/DDC accumulator width from 0Hz up to samplerate/2. For the common accumulator width of 32bits, the frequency resolution is better than 0.1Hz even for 400MHz samplerate.

[CJay]

that marconi sounds like it was designed for jamming radio signals! 

It was quite instructive to take a wander with a handheld to see just how far the sig gen in my spectrum analyser was audible for.

I'm a lot more careful with where I set it now.

[TechieTX]

How to spot the cheap-@ss Chinese clone chips:



If your chip looks more like the lower image, you got taken.  Don't rely on the datecode, as that changes when they want to.  The logo and font used on the real chip don't match the clone by a noticeable degree, and the clone is so far off-center that the logo is nearly off of the chip.