[Help] Rise/fall times too slow on nanosecond-wide pulses

[2sayle]

Hello everyone :D

I am currently using the STM32H743VIT6 (LQFP-100) on a custom PCB.

My goal is to generate fast positive pulses (ideal case would be a 10ns pulse) on TIM4 CH2 (PB8). APB2 is clocked at 200MHz so I can get 5ns ticks from my timer.

Generating 100ns wide pulses is quite easy using PWM by placing "Pulse" parameter on 16bits to 19 = (20-1) since 100ns/5ns = 20.

Problem is: if I want to narrow this pulse to 50ns for example, the trace of the signal starts to become rounded and doesn't reach 3.3V on top.

I suspect it is because of the rise/fall times which are about 30ns each.

Second problem is: When I change OSPEEDR register to VERY_HIGH, the rise/fall times are only reduced by a few nanoseconds while the datasheet shows rise/fall times less than 5ns for my range of capacitive load and supply (3.3V). I have estimated the capacitive load to be approx. 40pF.

Can someone enlighten me please ? :folded_hands:

2sayle

[Bob S]

Are you saying that when you output 100ns wide pulses the rise/fall times are shorter (i.e. sharper edges) than when you output 50ns wide pulses? That would be very strange.

Perhaps your estimate is wrong? Are the internal or external pull-up and pull-down resistors on that net? How about your decoupling capacitors on the VCC pins (this is a stretch)?

Since this is a QFP package, unsolder PB8 and bend it up off the pad and measure the rise time again. Resolder PB8 and (if you can) unsolder the load pin(s). Measure again.

[LCE]

Have you checked the OSPEEDR register contents?

Internal pull-up is off?

Are you 100% sure that there is no other load?

What about your measuring equipment, is actually the scope fast enough?

The probe calibrated / the compensation capacitance adjusted?

And the probe's GND is not connected to board GND with the typical few-cm GND cable?

[2sayle]

No, I would say that rise/fall times are constant (30ns approx) whether I program a 100ns or a 350ns time-width for the pulse, for example. But if I want to program a pulse under 50-60ns, the signal becomes rounded (and not squared anymore) : it's expected since rise/fall times are limiting the minimum width I can get.

The trace is not externally pulled since it is only passing through an analog demultiplexer and a SPDT switch. Internal pull-up seems to be off also when checking GPIO configuration.

GPIO_InitStruct.Pin = GTCH_TIM_OC_Pin;
GPIO_InitStruct.Mode = GPIO_MODE_AF_PP;
GPIO_InitStruct.Pull = GPIO_NOPULL;
GPIO_InitStruct.Speed = GPIO_SPEED_FREQ_VERY_HIGH;
GPIO_InitStruct.Alternate = GPIO_AF2_TIM4;
HAL_GPIO_Init(GTCH_TIM_OC_GPIO_Port, &GPIO_InitStruct);

I checked OSPEEDR register in debug mode and it was properly initialized.

I am not sure about the capacitive load of the signal but sure about the crossed components : it is CD74HC4067SM (through pin COM) and SN74LVC2G53 (through pin A).

Finally, the scope used is properly configured and calibrated. Alligator clip is connected to board's GND.

[AScha.3]

is your scope probe set to 10:1 ? and is minimum 100MHz spec. ?

otherwise you CANNOT test speed/rise time etc. !!!

[waclawek.jan]

> and is minimum 100MHz spec. ?

You can hardly see few-ns edges with 100MHz spec scope + probe... the significant harmonic content of a 5ns edge goes well up to few GHz range.

JW

[AScha.3]

i wrote : and is minimum 100MHz spec. ... because real hi speed probes and scope are not cheap - and not wide spread. i have 500MHz probes and a old Tek scope to test hi speed things, this can show 600ps rise time . but usually i take my DSO ("200MHz", more 120 in real) and a standard 100MHz probe , this can show 7ns rise at best.

typical for DSO is : Bandwidth x rise time = 0.45

(from : https://www.tek.com/en/support/faqs/how-bandwidth-related-rise-time-oscilloscopes )

[2sayle]

Effectively, I did not configured my probe in X10... I observe now a rise/fall time of 3ns approximatively.

My probe is Rigol RP1100 and datasheet says that I can measure down to 1.4ns rise times with X10 setting.

Now signal is better but I observe a lot of overshoot: is it due to a wrong configuration again ?

Thanks a lot for all the replies :beaming_face_with_smiling_eyes:

[AScha.3]

1. ok now - basically.
2. probe needs adjustment (square wave / overshoot , check with probe signal at scope)
3. hi speed testing only possible with 10:1 tip at point to test and ground at metal ring on tip.
4. line to tip gnd < 3cm from gnd on board (solder blank wire there).
5. with "standard" long gnd line /alligator clip signal never will be "perfect" looking

like this...

 

->

https://electronics.stackexchange.com/questions/491355/oscilloscope-probe-points

[LCE]

Oh boy, I didn't even expect someone to NOT use 10:1...

The GND probe point you see above should be available with each high speed probe / scope, if not, grab some silver wire and roll your own. :D

[2sayle]

Thank you all for your answers. Indeed, I am not used to probing on PCBs. Your advice has been invaluable! Do you have any manuals/tutorials that I can use to improve my hardware testing skills?