That thing is a 20 kW noise generator!

It’s the mid 1990’s and I am a design engineer for a company that designs and manufactures custom measurement systems and high-power electronics. Our products range from DSP modules for fiber optic current sensors used on high voltage transmission lines to ground power units for military aircraft. For more Tales from the Cube from my time at this company see my article “The start-up transistor works once!”

The local electric utility set up a research program to get better load impedance data for their distribution network. They contacted our company and explained their needs for a special test generator.

Do you have a memorable experience solving an engineering problem at work or in your spare time? Tell us your Tale

They have detailed 60 Hz impedance numbers, but they want wideband information. The measurement must be done live, i.e., with power applied. Without power, relays and contactors dropout and the loads are disconnected making the measurements almost useless. Also, many loads are non-linear: the current consumed is not linearly related to the applied voltage, so the operating voltage must be present during the tests to get meaningful data. The distribution lines they want to analyze carry three phase at 14.4 kV line to neutral at 60 Hz.

To get the data they need, they cannot change the power frequency: this would demand too much power and upset proper equipment operation. They cannot add a test sine wave onto the nominal 60 Hz sine wave, this would also require very high power and risk damaging or interfering with proper customer installations.

Instead, they chose a technique used in electroacoustics: Maximum-Length Sequence (MLS) [1] [2], also called pseudo-random sequence. This technique uses a number of flip-flops and a few exclusive OR gates to generate a series of bits. This test sequence is applied to the circuit and both current and voltage are measured. With appropriate data manipulation, the impedance can be found for a broad range of frequencies. The sequence of bits can also be generated with software.

Since the number of bits generated by the method is the longest that can be made with a set circuit, it is “maximum-length” (the ML in MLS). Once all the bits are generated, the sequence repeats exactly as before. In this case, instead of using the sequence as a test signal voltage, the sequence will turn a known load resistor on and off according to the bit value.

Since MLS is not well known, details are given in the sidebar below:

Inside MLS

In the time domain, the bit sequence exhibits pseudorandom characteristics. It is random if analyzed for series of bits shorter than its maximum of (2**n)-1 bits, where n is the number of flip-flops. It is pseudo because the sequence repeats exactly after generating the maximum number of bits and not all combinations of bits are generated. True randomness does not repeat.

The circuit uses a clock to trigger the flip-flops. It has a period of t seconds. The lowest frequency generated is fmin = 1 / t((2**n)-1), which is the inverse of the repeat period.

In the frequency domain, for an analysis bandwidth larger than fmin, the spectrum resembles white noise: equal energy per equal bandwidth. The spectrum is composed of a series sine waves, all whole number harmonics of fmin [3].

A pseudorandom generator circuit built with one exclusive OR gate and with 3 flip-flops clocked at 7 kHz generates a repeating sequence of seven bits. It’s repeat period is 1 ms. Of the 8 possible combinations of three consecutive bits, only one is missing from that sequence. In that sequence there is energy at six discrete frequencies from 1 kHz to 6 kHz before the first null at 7 kHz. All the frequencies have significant energy content.

The exact multiples of the clock frequency, 7 kHz, 14 kHz, 21 kHz…all have zero amplitude. This is caused by the ((sine x)/x)**2 factor for the power spectrum. There is energy at frequencies above the clock frequency but of lower amplitude, as given by the function.

A similar circuit with the same clock frequency and one more flip-flop will have its sequence of 15 bits repeat every 2.14 ms and generate 14 discrete frequencies from 467 Hz to 6.53 kHz each about 3.7 dB lower in amplitude than in the previous case. For a fixed bit voltage, the amplitude of the harmonics is proportional to the number of discrete frequencies produced. In this case, the number of frequencies between the nulls increases from 6 to 14, so the amplitude is lowered by the same amount 10(log(6/14)) or 3.7 dB each.

For a given output power or voltage and a given clock frequency, the finer the frequency resolution, the lower the power available at each frequency.

With a pseudorandom bit sequence as the input stimulation, it is possible to extract the system’s impulse response and from that, the system’s frequency response. By using a known load that is switched on or off according to the maximum-length sequence binary sequence bits value, the network impedance can be computed.

Since a single sequence is needed for the computations, there is no need to repeat the signal and the duration of interference with normal service is minimized. If desired, the test can be repeated to improve the measurement signal to noise ratio.

In this project, the electric utility will provide the main computer and processing software as well as the voltage and current transformers while we will provide the switchable resistor load bank. The test load will be operated at 600 V via a transformer.

They estimated that a 20 kW load would enable them to get reasonably accurate data. The line current at full load is nominally 11.1 A at 600 V.

The test system is to be easily transportable. Most of the equipment will be inside the electric utility truck. The three test transformers—one for power and one each for voltage and current measurement—will be installed onto the 600 V lines. Then the test resistor will be connected.

Due to the resistor size and dissipation, we decide to build it onto a two-wheel moving dolly also called a hand truck. This will make the test resistor easier to move about. A box will house the resistors, a fan, the power and control circuits, and the power supply for the electronics.

The electronics are made up of a six IGBT modules, six isolated gate drivers and some control, as well as the monitoring and interface circuit. The IGBTs will connect the load resistor to the line according to the sequence bit value. The gate drivers have under-voltage, over-voltage, and over-current fault detection. Each gate driver has two optoisolators: one for on-off command input and one for status (OK or fault) output. The data connection to and from the test truck is with fiber optic. To power the electronics and the fan motor, 120 V is taken from the utility truck 120 VAC utility outlet.

During the initial testing, I am asked to give a hand debugging the test resistor. Debugging is done after office hours so that we do not disturb the other tenant’s electric equipment.

Tony, the design engineer, tested the electronics and the resistors at low power with a DC power supply. When I met with him, he has just connected the oscilloscope power plug to an isolation transformer because our newest four-channel 200 MHz digital oscilloscope keeps resetting during tests. Powering the oscilloscope from the isolated output cures the resets. For safety reasons, the oscilloscope frame is still connected to the safety ground.

The test he is running is a low power 3 phase test with 120 V applied to the resistors instead of 600 V. This reduces power to about 800 W. After a few seconds something smells. We cut the power and investigate. We both conclude that what we smell is the odor of dust and oil that has accumulated on the power resistors and that we can safely continue testing.

A minute later a different odour appears. Again, we shut down the power. This time it is the varistor that protects the 120 V power to the electronics. It has overheated and is burnt. We decide to remove it and power the load resistor electronics from the same isolation transformer used for the oscilloscope. The 800 W test signal sent to the building electric system is coming back into the wall plugs and overloaded the varistor. Maybe we should have bought a UPS for this test.

We power up again. After a short time, the system stops because of a driver fault. The IGBT driver IC detects various faults, and all faults are reported into a single output pin. The six driver faults lines are sent through six individual optoisolators. All six optoisolator outputs are connected to one microcontroller input pin. We have no way to identify a specific IGBT.

Tony tells me he has tested the supply for the drivers, and they are safely within the voltage limits, they can drive the current needed to turn-on the IGBTs, and that there are no shorts in the power load assembly.

We continue testing and we get more faults, all the same as before. The faults take a few seconds to a minute or so to appear. It is the longest time the unit has run. Before this time, we had to shut it down because of other problems.

I ask Tony to show me the PCB artwork. He says he can do better, and he gives me an unassembled board. I look at the board layout and everything is good. I search for the gap between the low voltage electronics and the IGBT driver side where optoisolators sit and I can’t find it. It should be easy to spot 12 optoisolators. Finally, I spot it. I had missed it because there is a trace between the isolated sections! One long trace is set between the isolated areas of all the optoisolators. I ask Tony what signal that trace carries: he says it is the fault signal from the drivers.

I tell him I think capacitive coupling is causing the fake fault signals. I tell him we should cut the trace completely from the 6 fault pins and replace it temporarily with a wire placed far from the power section. We do that and test again. The unit runs for 20 minutes without a fault. This confirms it: the trace between the two isolated circuits picked up stray capacitive current from the high voltage side and when it is big enough it is detected as a fault. Removing the trace and replacing the missing connections with a temporary wire kept the unwanted transients from causing false error detection.

I tell Tony to permanently remove the copper trace so that the PCB will be able to carry the full voltage without arcing from the high voltage side to the trace and from the offending trace to the low voltage circuits.

The next day I ask the junior engineer assigned to the PCB design why he used the empty space to place that trace. He says he had one trace left to route and saw the empty space and used it. I tell him about optoisolators isolation, clearance and creepage distances requirements for PCB design [4] to make sure he does not repeat this error.

The unit was later delivered to the customer who used it to complete their project and had no problems with it.

The lessons learned are:

Make sure the PCB designers knows all the requirements for the PCB and know about isolation, clearance, and creepage distances as related to PCBs (IEC or UL 60950-1, section 2.10).
Implement rigorous design reviews not just of the schematic but also of the printed circuit board and use a list of items to check, always keep your list up to date, the list here is a good start [5].
Think of the PCB as a circuit element.
Make a test plan at the start and review it regularly, this will allow you to buy test equipment and design test procedures and set-ups before it is too late.
Know what overheated electronic parts smell like.
Have a clear overall view of the project and its impact on the electrical environment, clearly our test resistor is a 20 kW wideband electric noise generator!

Daniel Dufresne is a retired engineer and has designed digital integrated circuits, active loudspeakers, negative resistors and high power electronic converters. He also was a professor at Cegep de Saint-Laurent. He earned his bachelor’s degree from Ecole Polytechnique de Montreal. He lives in Montreal, Canada and still works on electronic projects and repairs electronic test equipment.

References

Article about maximum length sequences on Wikipedia. https://en.wikipedia.org/wiki/Maximum_length_sequence [Page retrieved 2024-01-15].
Jessie MacWilliams and Neil J. A. Sloane. Pseudo-Random Sequences and Arrays. Proceedings of the IEEE, Vol. 64, No 12, December 1976. Pages 1715-1729. It details the mathematical aspects of pseudo-random sequences and arrays.
Hewlett-Packard Journal, September 1967, Volume 19, number 1. Gives detailed information on the spectra of pseudo-binary sequences. http://hparchive.com/Journals/HPJ-1967-09.pdf [Page retrieved 2024-10-18]
A short video about creepage and clearance: What are creepage and clearance? (On line) https://training.ti.com/ti-precision-labs-isolation-what-are-creepage-and-clearance?cu=1135015 [Page retrieved 2024-10-15].
Wallace, Hank. Electronics Design Checklist, (On line) http://www.jldsystems.com/pdf/Electronics%20Design%20Checklist.pdf [Page retrieved 2024-10-15]. There should be items added in regards to isolation, clearance for optoisolators and trace size and current rating, see IPC-2221 and IPC-2152.
<!–

VIDEO AD

–><!–

div-gpt-ad-inread

–>

<!–
googletag.cmd.push(function() { googletag.display(‘div-gpt-ad-native’); });
–>

The post That thing is a 20 kW noise generator! appeared first on EDN.

Become a member

Become a member

Become a subscriber

Become a sponsor