ADL5303 input averaging for pulsed laser source

Hi Aubrey,

Thanks for looking into this issue. Here is our schematic:

We reverse voltage bias the PD in two modes: Low Bias = 0.65V and High Bias = 4V. The high bias is essential so that the PD can experience a high optical power input without being forward biased. The PD becomes forward biased at I_avg = (4.0 − 0.5) V / 222.85 kΩ ≈ 15.7 µA. 

Our goal is to measure up to 25 µA average of photocurrent which will require a future redesign. This will allow us to measure 15.9 mW optical peaks at ~0.5% fine duty.

The problem isn't that ADL5303 itself is too slow. The problem is with the input filtering which we require as we need to average the pulses at the current stage to derive the arithmetic mean. Your example circuit doesn't do this averaging which would require us to then average the output VLOG(t) instead. Averaging in the log space would result in computing the geometric mean of the pulsed photocurrent instead of the arithmetic mean (true average power). This method results in underreporting of the average power by many decades which would not be correctable via a fixed calibration curve. 

Since posting this question, my team is no longer concerned with the dead-time (LPF charge up time) as this is a one time occurance upon power up. Instead we are mostly concerned with this Post-step tail. Essentially, we are seeing a steady-state creep (which is logarithmic) after a step response in optical light. This creep settles in ~5min and we believe it is likely due to the input LPF that we have added. 

Swapping the photodiode with a higher current version doesn't solve our problem as it reduces the max measurable current before the photodiode becomes forward biased. 

I also noticed that the ADL5303 datasheet shows the bandwidth collapsing to roughly 2 kHz at 1 nA. Does the LTSpice macromodel you used capture this current dependent bandwidth roll of?