FFT Adaptive Filtering: LMS Algorithm resoant frequency locking / General Science and Electronics / Forums

Forums

4hv.org :: Forums :: General Science and Electronics

« Previous topic | Next topic »

FFT Adaptive Filtering: LMS Algorithm resoant frequency locking

Move Thread

LAN_403

Danielle

Sun May 31 2009, 07:27PM

Registered Member #632 Joined: Mon Apr 09 2007, 01:09AM
Location:
Posts: 267

I was thinking about methods other than usual for high power industrial resonant inverters, and I thought of the possibility of looking for the ripple out of an injected PWM and filter this adaptively to get a frequency lock. Aside from the DSP speed issues and impulse response and the necessity of a few cycles to be monitored in order to get an accurate FFT on a signal, I think this could be the future of resonant DSP inverters.

I wrote up a PDF explaining how it would work.

I still need to write out a block diagram and the final coding, but I just wanted to see what people though.

and here is a link to adaptive filtering algorithms:

]resonant_frequency_fft_tracking_metho1.pdf[/file]

GeordieBoy

Sun May 31 2009, 10:21PM

Registered Member #1232 Joined: Wed Jan 16 2008, 10:53PM
Location: Doon tha Toon!
Posts: 881

Having experience in both DSP and power electronics disciplines, I'm struggling to see what you are getting at here?

What is the application? Are you suggesting using an FFT to track the resonant frequency of a Tesla Coil whilst it is being driven by an inverter? Or are you suggesting this technique for other applications like the SLR inverter?

In order to perform an FFT on the current waveform of a typical SSTC you would have to sample at somewhere near 500kHz or above depending on the harmonic content of the current waveform. Then you need to sample for some considerable time because the frequency resolution of the FFT bins is inversely proportional to the length of the time domain input. There's no point doing an FFT over only a couple of cycles because the frequency resolution will be very coarse.

In case you are not aware the current state of the art in frequency tracking SSTCs uses an analogue phase-locked loop. This provides good frequency agility with customisable lock-in range and response time. It also does this in the continuous time domain with cheap analogue electronics, zero latency and almost infinite precision! Any FFT based scheme is ultimately going to have a limited frequency resolution ability, and significant processing latency.

By all means use white-noise excitation and an FFT to find the initial resonant frequency of a TC in the measurement stage, but it sounds overly complex and expensive for realtime control.

Just my initial thoughts,

-Richie,

Steve Conner

Sun May 31 2009, 10:37PM

Registered Member #30 Joined: Fri Feb 03 2006, 10:52AM
Location: Glasgow, Scotland
Posts: 6706

All I'm going to say is that I was programming DSP and micros for a day job when I built my first PLL SSTC drivers. Around this time, Terry Fritz was really keen on the idea of building a DSP driver (DSPRSSTC?) but I just couldn't see how any amount of DSP could beat that little $0.79 4046 chip, for the reason Richie mentioned: latency. However, I believe induction heaters do use similar algorithms, just without the FFT part.

Now I'd probably be trying to get the whole lot onto a FPGA, with an embedded 8051 core to control the USB cup warmer.

Danielle

Sun May 31 2009, 11:19PM

Registered Member #632 Joined: Mon Apr 09 2007, 01:09AM
Location:
Posts: 267

Thanks for the feedback, I was thinking that this could be a reasonable method for tracking resonance in a series-loaded resonant inverter (SLR) but your right that this would be over complicated for any application, and the fact that its response is non linear, delayed, and overall an expensive system, its not feasible.

Sorry for bothering with that, I guess even though the PLL is susceptible to some noise and delay issues, it is the best method for resonance tracking.

Danielle

GeordieBoy

Mon Jun 01 2009, 12:03PM

Registered Member #1232 Joined: Wed Jan 16 2008, 10:53PM
Location: Doon tha Toon!
Posts: 881

Steve: I think modern solid-state Induction heaters use micro control to track the resonant peak for several reasons. It allows them to adapt the control parameters to different situations, like when someone changes the work coil or tap on the matching inductor, or the workpiece goes through Curie temperature. It also allows them to track the maximium power point up to the desired power setting etc where there are multiple constraints to the operating envelope. Finally, if the micro is already in control of power-throughput, drive frequency etc, it makes metering these quantities trivial. I guess it's also easier to hook equipment into Industrial controiller busses like CAN too if there's a microcontroller in there running the show.

Danielle: In some circumstances even the PLL can be considered as "over complicated". When the gain and phase responses of the system are favourable you sometimes only need to provide negative feedback and sufficient gain to make the system oscillate at its natural resonant frequency.

This "power oscillator" is what people have been using for years to track the resonant frequency of quartz crystals in oscillators etc. The same idea is used in most DRSSTCs where the inverter output current sensed, amplified, and fed back to control the switching instants of the IGBTs.

This works well except for mode-hopping and the inherent delay in the feedback path which pulls the system slightly off resonance.

These two areas are where the PLL has the advantage. It can be adjusted to compensate for propagation delays (phase shifts) in the power electronics, and it can be set up to limit the locking range to prevent mode-hopping if desired.

If I really wanted to make a discrete time controller for a DRSSTC i'd probably make a digital implementation of Steve's PLL controller. But for the reasons discussed I still don't think it is a good use of time and money.

-Richie,

Moderator(s): Chris Russell, Noelle, Alex, Tesladownunder, Dave Marshall, Dave Billington, Bjørn, Steve Conner, Wolfram, Kizmo, Mads Barnkob