Hi Rujun,

just a quick consideration on a related topic: I am working on simulation and model validation, focusing on electric networks and in particular railway traction simulation.

One of the problems once you have experimental data to compare, is how to compare them: it much depends on your waveforms, spectra, etc. and their characteristics, plus the level of statistical confidence you want to achieve. We consider peaks and valleys of resonance/antiresonance behaviour, curve slopes between them, phase jumps and slopes, and of course the amplitude! Not all performance indexes behave in the same way, especially if they come from different fields (economics, crystallography, etc.).

[links added ... hope they work; here you can find further references to the general issue of validation of simulation, but with elements that may be applied in a broader scope]

Sorry for being slightly off-topic, but evaluating performance of an instrument r a simulation model, when the output is complex, is not trivial.

cheers

andrea

Article Evaluation of performances of indexes used for validation of...

Chapter Performance of indexes used for model validation

Article On the Validation of Models of Large Complex Electrical Systems

Dear Andrea, 

I spend some time to read your papers. I think your idea is very important to the evaluation of TEM instrument system. We are developing a new TEM system and making testing everyday. There are a transmitter and a receiver in a typical TEM system. The load of transmitter is an induction coil or a loop. It's impossible to transmit a perfect rectangular waveform for this kind of load. There exist a limited resonance frequency for a loop. There exist same problem for the sensor of TEM receiver. The sensor have limited bandwidth. There exist distortion for the received waveform. In general speaking, we can not transmit a perfect waveform, and we can not receive a signal in perfect way. However, Theoretical study is based on perfect signal transmitted and received.  Our goal is to try our best to transmit and receive signal in best way. Therefor, we need a method to evaluate the shape of transmitted waveform and received waveform. 

Best wishes,

Rujun

Thanks Rujun! Well, a pure sinusoid can come out from a filter (you put a resonating LC before your coils, maybe decoupled by some resistive network, otherwise the "talk together"), or using a pure generator (if you don't need a lot of power, a RF generator, a signal generator with extended bandwidth, or a VNA, or spectrum analyzer with its tracking generator).

It depends on several details ... if you like to write me and send some description, it's welcome! name dot surname at gmail dot com.

Hi Andrea! Thank you very much! The transmitted waveform is a rectangular waveform with duty cycle as 50%. The attachments are papers about TEM. It's a challenge task to produce a very good rectangular waveform on inductive load. Your experience may inject new oil in the instrumentation of TEM. 

Hi Rujun,

I studied the papers you sent me (nice things, I didn't know!) and the quality of waveform depends of course on the amplifier you use and the frequency (or better the rise time and the period) you want to push: the larger the "frequency2, the larger the voltage and the chance that you reach the Vrail of the amplifier.

In any case you need a current source amplifier.

- The bandwidth for the most challenging waveforms are around 5 to 20 us of on-time, let's say about some hundreds kHz of bandwidth.

- The inductance of a 12mx12m 4-turns coil is large (but that is supplied with a slower waveform, if I understand well), while a smaller one (e.g. less area, or less turns) is supplied with the fast waveform I mention above. L = 80 uH for 1 turn and 12x12 coil, = 1.3 mH for 4 turns and ideal coupling.

So, at 100 kHz you have XL=50 ohm for the smaller one.

If, as said, you want to inject 20 A (or even more) that's a fairly high voltage.

Presently I am concluding the pre-series of some power amps that shall go to certification (September?) and they are able to do the following:

1) 20 A x 50 V, or 2) 10 A x 100 V with a bandwidth of about 800 kHz, voltage and current source (switchable) in the same unit.

Units can be paralleled, but I didn't try to connect them in series to increase the voltage: an output coupling transformer to parallel as many as you want was designed but for the moment is not in production (1 thing at a time!).

In any case a larger amplifier is possible up to 6U height, that is around 3 kW. Above that size probably it is better to connect units together.

The transformer of course compels you to say goodbye to dc component, but "dc" concept is fishy: if your waveforms last for ms, then that's some hundreds Hz, not dc, and with the largest transformer (large core, large Voltxsecond before saturation) you should be able to go down to about 10 ms without problems at maximum current).

Guessed the whole stuff almost right? not to bore the audience with details, if you like to send me waveforms, or specifications, or whatever you can write at name dot surname at gmail.

cheers

andrea

Just a second thought.

Processing of received signals shouldn't be tremendously difficult: I think that you identify time constants, even if the papers you sent are not describing how they processed the results.

In any case the results are quite interesting for many uses: I sometimes look for soil resistivity values for lightning protection and stray current simulation.

Sorry for insisting :-)

Of course if you do not want a fully controllable waveform but an approximate square wave with also dead times at zero value in between positive and negative "half cycles", the straightforward solution is that of an H-bridge between a +Vdc and -Vdc using MOS or IGBTs able to go up to a thousand Volts with no problem (the MOS yes, honestly, the IGBTs not) + an optocoupled or transformer coupled driver (isolated drivers are commercial).Of course there will be the L/R knee in the waveform at the rising and falling edge (seen also in the papers you attached, so it shouldn't be a major problem) and some risk of overvoltages if the inductor is left floating (in this case abundance of free wheeling diodes and/or voltage limiting devices ... I prefer the former whenever I can use them).

For IGBTs of 100 A IC switching is well sub-us and you get what you want. Only to pay attention to triggering of parasitics and ringing.

Dear Andrea Mariscotti,

We usually use IGBT or MOS to build an inverter converting DC to AC. The inductance and capacitance in the loop is the major problem for linear and short turn-off time. It is the most important problem in TEM transmitter. I attach some popular circuits used in TEM transmitter. The circuits include inductance of loop for simulation. I hope that you have new idea for this problem. 

Dear Andrea Mariscotti,

This is the reply for your answer about received signal processing.

There are some difficult problems in received TEM signal processing. The received TEM signal ranges from several volts to several nV.It's very weak at late time because it decays rapidly. There are many interferences inside a TEM signal, such as sferics, VLF radio, power line, and other human made EM interference. Robust stacking is usually in TEM signal processing. We are seeking better method to process TEM signal. TEM is very usful in metal finding, hydrogeology, mineral exploration, and so on. If the depth of exploration is less than 20 m, resistivity method (ERT) is better than TEM. ERT can make 2D/3D imaging of soil resistivity.