Saw tooth signal when measuring pulses at high input impedance

[Guille]

Hi everyone,

I'm working at a LASER lab measuring short light pulses with a photodiode (PD). The setup is simple: A LASER shines a train of light pulses into the PD, each pulse lasts for 50fs and the next pulse arrives to the PD 0.2ms later (so something like a PWM signal at 5KHz with a 2.5E-10 duty cycle), the PD is reverse biased with an internal battery (manual linked below) so it should put out a short current spike for every light pulse, to read it we plug the PD into an osci (using a 50Ω coax, 2m, RG58 cable) and use the osci's input impedance as a termination resistor to read the current spikes.

As expected there are some differences in the signals' shape when setting input impedance to 50Ω or 1MΩ, and yeah I'd expect there to be some LP filtering when combining junction and cable capacitance with a big termination resistor but not like this:

Fig. 1: 50Ω input impedance. Yellow trace, pulses marked with vertical cursors for visibility

Fig. 2: 1MΩ input impedance. Yellow trace.

First image shows the waveform for a 50Ω input impedance and second one for a 1MΩ, as expected the second one has a higher level at the expense of a super long decay time but what rubs me wrong is the rise time is almost instant in comparison! I would expect a low pass filter to be independent of "direction" (i.e. rising or falling). What do you think? My guess is the photodiode is playing some role into this, kind of how an Attack/Release synthesizer module uses diodes to discharge a capacitor through different resistors but the specifics are lost on me.

Anyway my team and I would appreciate any insight into this phenomenon and how to solve it since in the future we will have to use 10m long cables and I fear cable capacitance will worsen the problem.

[EC8010]

Read the manual for the oscilloscope; I've been caught by this one. Oscilloscope bandwidth is often deliberately limited on the most sensitive ranges in order to keep the noise down and it's buried in the manual's small print. I ended up putting a sticker on the front of a Tek DPO4034 to remind me of bandwidth on the three most sensitive ranges. Later Teks tell you what the bandwidth is at all times.

You need a fast charge amplifier near your detector (<200mm away). Ortec and Canberra/Miriom make such things or you can design/make your own from scratch. Also, you appear to be sampling at only 10MS/s. Choose a faster time base that allows faster sample rate. Oh, and read the manual for how to get a screen grab onto a USB stick rather than pointing a 'phone at the screen.

[David Hess]

You are not seeing any low-pass filtering here.  For that the timebase will have to be set much faster so that you can see the rise time.

In the 50 ohm case, the photodiode sees 50 ohms at all frequencies and there are no reflections except for the mismatch caused by the oscilloscope's input capacitance, but this may not be significant.

In the 1 megohm case, reflections in the cable limit rise time, which does not show here with the slow timebase, and the cable capacitance plus the oscilloscope's input capacitance act as a capacitor accepting charge from the photodiode.  The 1 megohm impedance slowly discharges this capacitance.

2 meters of cable is about 10 nanoseconds and 180 picofarads.  If you want to see the reflections producing the rise time, the oscilloscope sweep will need to be like 50 nanoseconds/division or faster.

[tszaboo]

What David wrote is quite true, but I prefer to look at it another way. Photodiodes are current sources, that's why they use transimpedance amplifiers with them (that turn these currents into voltage). When you terminate it with 50 Ohm, there is something to turn this into voltage. When you have the 1Mohm scope probe, you just dump current into the scope probe, and then quickly make the diode open circuit, and let that charge float, and get slowly discharged by the oscilloscope.
Which is nice when you want to invent DRAM, not very useful otherwise.

[Guille]

I definitely see how reflections could cause headaches here with such a mismatched input impedance but after doing a quick simulation in LT-Spice I'm not fully convinced this is what we are seeing here.

In the simulation (linked below) I purposefully ignored the transmission line or parasitic effects on the osci, to see if this "saw-tooth" effect could be accounted solely by an uneven charge and discharge cycle in the photodiode's capacitance, I used the diode model stated on the manual (linked on main comment) and I think the simulation achieves a similar result to what we see on the real traces.

Fig. 1: Voltage measurement at osci for 50Ω input impedance

Fig. 2: Zoomed in version of previous measurement, a longer decay is present but short

Fig. 3: Voltage measurement at osci for 1MΩ input impedance. Decay is much much longer

If you can't see the figures have attached them at this imgur post: https://imgur.com/gallery/saw-tooth-connundrum-zowqxrL

Reflections are not being simulated here however we can still see this short rise time and long decay for 1M (And something similar alebit much shorter for the 50Ω). I'm wondering if the saw-tooth effect is mostly due to how the diode could charge and discharge the input capacitor at faster or slower rates depending on whether it's rising or falling.

[David Hess]

I definitely see how reflections could cause headaches here with such a mismatched input impedance but after doing a quick simulation in LT-Spice I'm not fully convinced this is what we are seeing here.

We are not seeing the reflections because the oscilloscope sweep is way too slow.  The reflections are over in less than 100 nanoseconds.  The reflections produce the same exponential curve associated with a low pass filter, but it is not visible at such a slow oscilloscope sweep rate.  With a slow view, the unterminated transmission line looks like a capacitance of about 180 picofarads.