[Blocking output] Bouncing right after falling edge
A strange behaviour has been detected in the output Blocking stage:
The Blocking driver is showing a bouncing right after the falling edge.
MEASUREMENTS: TIPS AND SET-UP*
Measuring this effect is a little bit tricky. Bad schema of connections can produce measurements that do not correlate with what will happen in the circuit when the probe is not connected.
Firstly, we have to keep in mind the isolated coupled inductors, MSD1278-104ML. This coupled inductors offer 500V coil-to-coil isolation. One of the leads of the coupled inductors is connected to ground and the other one is not connected to it to preserve some isolation. Naturally, some parasitic capacitance will appear in the coil that is isolated from the board ground. This parasitic capacitance will produce a virtual grounding at higher frequencies. However, at lower frequencies (like the ones responsible of the top "prairie") the isolation is more effective.
Keeping in mind that, isolation must be preserved while measuring the shape of the pulses to avoid misleading reads in the oscilloscope. Because of this, while measuring the isolated part of the circuit (that is the second coupled inductor of the blocking driver) an isolated oscilloscope should be used. I spent three hours seeing wrong results due to the fact that I was connecting a non isolated oscilloscope to the output blocking driver. Result: "ground bounces" in 3V3 supply rails due to a non-expected ground loop.
Once a correct set-up of the measurement system is configured, we can go ahead with the definition of the issue.
> Set-up
> The instrumentation material used is:
>
> * Agilent 33250A pulse generator. A pulse of 1us, 1Hz cycle, 5ns
rise, 3V3 top is set.
> * Tektronix TPS2024: isolated oscilloscope
> * LeCroy WaveJet 324A: non isolated oscilloscope.
> * Tek P6134C: probes used with both oscilloscopes. 10x attenuation,
10MOhm, 10.5pF.
> * 50 Ohm cables: different lengths of coaxials with different
cross-sections. Hence, slightly different capacitance per unit length.
> * A 50 Ohm termination and several "Ts" for interconnection.
> * 8 channel Blocking repeater: to check if the bouncing is producing
glitches.
> Results
> # When the cable is short, little or no bouncing is observed.
> # When the cable is longer, a bouncing is observed. The longer the
cable, the slower the bouncing. The frequency of the bouncing ranges
from 3 MHz to 1 MHz, with the cables available in the lab up-to-day. The
height of the first bounce can go to 1.6V (close to threshold in legacy
boards).
EXPLANATION*
At the begging it seemed a problem of fast falling edge and ringing in the line. Reducing the slope of the edge could help. This solution was tried out and, in fact, some of the bounce was reduced but a big quantity was still there. However, if the circuit is under-damped modifying the value of the resistor could make it work. It was obviously affecting to the bouncing but not addressing the expected result.
Then, in RLC circuits, the damping is related with the inductance and the capacitance of the rest of components. In our case, the isolated part of the coupled inductors forms a RLC circuit in which:
- R is the parallel of the snubber resistors and the 50Ohm terminated line
- L is the inductance of the second winding of MSD1278_104ML
- C is the capacitance of the cable, at a given length.
Analyzing the RLC parallel tank, with cable distances of around 10mts, reveals a resonant frequency around 1-2 MHz, which is the same as seen in the bouncing. So, it is possible that the source of the problem lies in that resonance.
SOLUTIONS TO BE PROVED*
Subsequently, as we cannot act over C and little we can do to R, the inductance L has to pay-off. When the initial design was done, the value of the inductance was targeted for being able to be compliant in the widest set of operation. That means repeating pulses of considerable length (4us for a really small quantity of legacy boards). That means that the magnetizing current should be out of saturation, which was accomplished by MSD1278_104ML (1:1 coupling with 0.98 coefficient and 100uH of primary inductance). As the resonant frequency is inversely square related with inductance, if we want to shift a decade up the resonance frequency (which, indeed, will be later attenuated by the transmission line) a reduction in a factor of 100 is needed. MSD1278_472ML (1:1 coupling with 0.98 coefficient and 4.7uH of primary inductance, 16A of saturation current) will do the job for this task. It should be noted that due to the lesser value of the parasitic capacitance of the MSD1278_472ML, sharper pulses will be achieved, so better decoupling will be probably needed.
Several models have been already requested:
- MSD1278_472ML
- MSD1278_562ML (shipped on 6 June)
- MSD1278_103ML (shipped on 6 June)
someone of these parts will be replacing MSD1278_104ML, depending upon best availability.
And a long 100mts cable has been requested to perform a signal quality test.
Ever should be in the lab by the end of the week, begging of the next one.
WHY THIS WAS NOT DETECTED BEFORE*
> Protoboards
>
> The connections were not that long enough to see the bouncing.
> Simulations
>
> Simulations models are difficult to be configured in such a way that
all the parasitics are consider. Let's show the case of the coupled
inductor. The manufacturer is providing the values of the coupling
coefficient, parasitic inductance, primary inductance, internal
resistance and SRF. In that case, we have to extract the SRF upon the
previous data. Then, it comes board parasitics like inductance due to
vias, coupling between the transformer leads, lead to ground
capacitance...
>
> Furthermore, there are internal issues in simulators that are not
working that well. MOSFET simulations are not accurate depending on the
version of SPICE used underneath the CAD. And, even in later versions of
SPICE, some models (like the "simple" BSH103 N-channel MOSFET used in
this project) are not behaving as they are supposed to (BSH103 is not
showing the double-plateau profile, typical from MOSFETs while charging
up).