Paul and Sabrina's Cheap 3 Phase Inverter (AC Controller) with Field Oriented Control

[Astro]

I don't have a resolver on my motor but it has been interesting reading all the discussion on getting them to work with Paul's controller.

Is there a reason why something like the Analog Devices AD2S90AP couldn't be added to do all the interfacing and maths?

They are pretty cheap on eBay. A quick search brought up several listings for less than US20 shipped.

1pcs Resolver Digital Converter IC Analog Devices PLCC 20 AD2S90AP AD2S90APZ | eBay

1pcs AD2S90AP LOW Cost Complete 12 BIT Resolver TO Digital Converter | eBay

[e*clipse]

To be perfectly honest, no. In fact, unless some trickiness is employed, I'm not sure the digital output can't be more quickly interpreted. Also, the Tamagawa version is used by Toyota - they have two in every Prius ( motor, generator ) So, the reliability of the concept and that IC is well proven.

My design should be cheaper, because both will need a custom PC board - that cost will be the same. Differences in circuit component cost are the main issues.

Actually, I found this out about 3/4 of the way through the AD app note: It makes perfect sense, as the IC has no power connections or heat sinks. They both need the power circuitry to drive the resolver. They also need the power supply to drive that amp. In other words, about 1/2 of the cost of the circuit that I am working on is that driver. It is the reason for the previous post. And - it doesn't go away with the AD chip or the Tamagawa chip; you're just confined to 10kHz or maybe 20kHz.

The other device that's a major cost in my previous design is the analog switch. The price of this comes down dramatically when it's spec'd for sub 5V, always positive. Power for this IC also becomes much easier.

So, in the end I have to deal with the same problems - just understanding the core nature of the problems is helping provide a much less expensive solution.

[Astro]

In the application note AN-234 it shows a circuit diagram with the outputs of the resolver directly connected to the AD2S90 chip. For the input to the resolver the circuit has a couple of op-amps on the input to the resolver (ref). The op-amps are both on a single US$3 AD712 chip.

The other thing that i thought would make this chip useful was the 1024 line incremental encoder emulation. Wouldn't this be able to feed directly to the encoder input on Paul's controller?

[e*clipse]

Check out the Tamagawa app note for a more in-depth explanation of the situation. This is what actually needs to be done for an automotive environment.

What I've found from actually building these things is that on one hand, you can build something that just works in the lab. It can be very simple. Now put it in a noisy environment (and an electric motor is the definition of a noisy environment) and see how the simple circuit works. Remember, at high frequencies, the wires to the resolver are excellent antennas. Also, the PWM power supply that's supplying that +/- 7 volts for the op-amp (plan on SOME signal loss) can also be a very significant noise source. Attempting to do precise instrumentation of subtle voltages with no filtration or shielding will lead to lots of grey hair.

[Astro]

Originally Posted by e*clipse

Check out the Tamagawa app note for a more in-depth explanation of the situation. This is what actually needs to be done for an automotive environment....

No worries, just thought i would mention it in case an easier solution was being overlooked.

I can understand the noise issue. It was something that was high on my list of reasons for using an encoder back when Paul asked for opinions on resolver versus encoder.

Its a pity that you can't just swap out the resolver for an encoder. Maybe an encoder bearing?

But when you get it working it will be a real neat solution. Bolt the MGR to the chassis, connect the half shafts, connect a few cables and away you go. No adaptor plates or input shaft couplers. And at the cost you could have a spare on the shelf ready for a quick swap over if the current one developed issues.

[e*clipse]

Astro

You're completely right - we should be looking for clean, easy solutions.

Thank you. Really. I guess I need to be reminded of this from time to time.

-E*clipse

Originally Posted by Astro

No worries, just thought i would mention it in case an easier solution was being overlooked.

[MPaulHolmes]

You know when I said I understood everything finally? I was wrong! But now I do! haha. Well, more or less. I found a mistake in the SVM step. I had Va, Vb, and Vc, (all of which could have been nonzero), but then I was using 2 of the component vectors as if the 3rd was zero. The very final step of the field oriented control was to find an equivalent vector, but only using 2 of the 3 unit vectors in the directions from Va, Vb, Vc. I downloaded tcc, (tiny c compiler) and have debugged each of the steps of the FOC code, with lots of printf outputs, and graphs in excel. It's all actually working! I was wondering how I was going to debug it. It would have been tough without a separate c compiler.

[thingstodo]

Originally Posted by MPaulHolmes

I was wondering how I was going to debug it. It would have been tough without a separate c compiler.

Great Idea! The C code was generic enough that it ran on two different compilers? That's a feat in itself!

[e*clipse]

Woo Woo! I'm making some progress.

I got the sine wave I was looking for a couple of nights ago - 9V p-p @ 50kHz. The only problem was the power op-amps I am using (those ST L-165 obsolete ones) were getting awfully hot.

So, I decided, what the heck, try it anyway, just monitor the temperature. My lab was 33.1C (yes, really it's pretty hot here) and the temperature of the power amps just tied to the 'scope was running 55C. A discussion about this heat problem led me to check the load impedance. Myprobes resistance is about 245Ohms; I've heard the capacitance is significant, but I have no way to check it. So, knowing that the impedance of the resolver is 1152 Ohms at 50kHz ( I calculated that previously ) I felt it was safe to plug the circuit into the resolver.

Well, the good news is that the signal is VERY stable, and did not attenuate at all due to being plugged into the resolver. The power op-amps heated up to about 60C while driving the resolver. The output is exactly what is predicted by the transformer ratio, or 2.2V p-p.

Done with that, right? Well, not sooo fast....

1) I'm very unhappy with the hot power amp. It should not overheat (yes, I think it's overheating) with no load. Perhaps I should not focus on a solution for this amp, as I'm not going to use it in the real circuit anyway.
2) There is a LOT of attenuation in the circuit before the power amp. The input is 5V p-p, and the signal the power amp finally sees is about 0.35V p-p. Is that a big deal? Considering op-amps can support gains over 100, perhaps the gain of 13 isn't. Also, highly amplified low amplitude signals are big noise problems. The low output amplitude seems weird, so I looked into that a bit.

I guess I could add a point to the "why bother when you can buy one" list for Astro. Sometimes the best way to force yourself to learn something is to do a project.

Here's what I learned - there are hundreds of op-amps out there, all with different specs that vary subtly from on another. In this case, I was using two different CMOS op-amps from Microchip, the MCP602 and the MCP6002. Both are low power, high speed precision op-amps intended for instrumentation. The MCP6002 is a newer version, with lower power consumption and a slightly worse Gain-Bandwith product. The MCP6002 GBwP is 1MHz, while the MCP602's GBwP is 2.8MHz. I figured that for this circuit, I just wanted a gain of 1 from each stage, since I was primarily doing waveform shaping. A Bandwidth of 1Mhz should be fine for a circuit needing a 50kHz signal, right?

NO. Not even close. The MCP6002 (which I was using exclusively for this circuit) cannot do the simple task of buffering the 50kHz input signal. It puts out a sawtooth (triangle) waveform with a peak of about 75% of the input. It also won't follow a sine wave above about 30kHz; there's lots of signal lag, leading to significant phase lag between the input and output. WTF?? It turns out that you have to look at a number of other specs as well. In this case the MCP6002 has a terrible slew rate - 0.6V/us. This means that it can barely track the square wave at pedestrian speeds below 10kHz. This also means that I basically built a circuit to make a sine wave out of a triangle wave, and that the waveform distortion DUE TO THE OP-AMP was significant enough to help create a sine wave. Meanwhile, the MCP602 has no problem accurately tracking the 50kHz square wave. Its slew rate? 2.3V/us. Meanwhile the power amp's slew rate is 8V/us.

Back to the drawing board. I really don't want the success of something to be due to a major screw-up. Well, it was a good lesson...

[MPaulHolmes]

Rule number 1: Things never work like you want them to (especially at first).
Rule number 2: Just when you get used to rule number 1, gosh darn it, something works. haha.

It sounds like you are getting REALLY CLOSE!!!