Please excuse the awkward phrasing of the question; it seems that LEDs have pulse latencies of nanosecond and sub-nanosecond durations. The question is, how was it possible to measure such precise increments of time? Is there ultra-high frame rate footage of this?

  • \$\begingroup\$ What are you really trying to do, and how well do you need to know the latency? Generally if you're designing something with pulse widths or delays below 5 ns or so, you try to design it so the exact value of the delay doesn't matter. \$\endgroup\$ – The Photon Oct 30 '13 at 0:01
  • \$\begingroup\$ This question came up in a comment to electronics.stackexchange.com/questions/86717/…. It is a theoretical question. \$\endgroup\$ – RJR Oct 30 '13 at 0:56
  • \$\begingroup\$ Yes, this is more of a hypothetical question regarding how manufacturers actually determine such latencies. \$\endgroup\$ – ayane_m Oct 30 '13 at 3:56
  • 1
    \$\begingroup\$ Not really hypothetical: It's commonly required to determine maximum usable signal frequency of an LED or laser diode, to allow binning for different applications. Manufacturing variations are high enough that a general datasheet or manufacturer's info is not good enough for determining actual maximum data-rate. \$\endgroup\$ – Anindo Ghosh Oct 30 '13 at 9:47

An inexpensive method of measuring rise and fall time limitations of an arbitrary waveform, is to start with a square wave of a moderate frequency, and then systematically increase the frequency while keeping duty cycle constant at 50%.

The average intensity of emitted light is easily measured, even by using something as basic as a CdS light-dependent resistor (LDR) cell.

As the switching frequency increases, rise and fall slopes become dominant factors in intensity of resultant signal, as illustrated in the graph below:

Graph of LED signal intensity

Note that the rising slope, and separately the falling slope, are nearly identical for signals of 50 through 200 MHz. What changes is the amount of time per cycle the signal stays high, or low. At 200 MHz, the LED intensity never reaches the plateau at all.

  • For very low frequencies, the average intensity is reasonably close to 50%, dominated by the "on" plateau and the "off" plateau.
  • As frequency rises, the sloped edges take up a significant part of each time cycle, so average sensed intensity begins to drop.
  • Once the frequency hits a level where the LED cannot fully turn on at all, the sensed average intensity drops much faster.

In the experiment from which the graph was taken (the paper is not publicly accessible), the average measured intensities were reported as:

  • 49.125% at 50 MHz
  • 43% at 100 MHz
  • 31.6% at 200 MHz (note the drastic intensity drop)

Thus, with fairly low tech, non-exotic means, the sum of LED rise and fall times can be determined.

To distinguish between the rise and the fall time values, the same exercise is repeated with different duty cycles, alternately minimizing the "on" plateau, and the "off" plateau to insignificance. Thus, the contribution and thereby the duration of each of the edges can be determined. I don't really understand the math of this last bit, so I'd leave it to someone else to explain it.

  • \$\begingroup\$ Wow, that's ingenious. For the math part, perhaps it's the oscillating integer function or some similar transforms. Has this ever been captured on camera though? using some thing like in this video? \$\endgroup\$ – ayane_m Oct 31 '13 at 3:28
  • \$\begingroup\$ Thanks, @abckookooman. Ingenious, inexpensive solutions are crucial in a country like mine, where throwing thousands of dollars at a problem simply is not an option. :-) Best of all, the end result is often easier to obtain and more reliable, as one avoids the problems of keeping a more delicate, fancy piece of equipment in perfect calibration. \$\endgroup\$ – Anindo Ghosh Oct 31 '13 at 5:44

It is possible to measure optical pulses of LEDs and Lasers to picosecond resolution, and it is something I do regularly as an electronics/photonics person. The trick is to work in the time domain.

To give an example of this, I have seen photodiodes exhibiting bandwidths of GHz, 40 GHz is quoted by a previous answer. However, I would imagine such a detector being quite expensive, and secondly you would need an oscilloscope of a matching bandwidth to record the resultant waveform for analysis (which would be horrendously expensive, I have only ever seen such scopes loaned on a day by day basis!).

Time to Digital Converters (TDC) have resolutions of 100ps or less, with modern low cost devices typically having ~30ps resolution. Single photon detectors often have electrical resolutions of a similar order, hence time measurements are feasible (without silly budgets!) to ~30ps. A technique known as Time Correlated Single Photon Counting (TCSPC) can then be utilised to measure the optical pulse of the LED to a resolution determined by the experimental setup. TCSPC is essentially just the histogramming of time differences between two TDC channels. I.e. measure the time difference between two signals a million or so times, and form a histogram of time differences.

The process is as follows:

  1. Drive the LED from a pulsed source. Utilise the trigger for the pulse source as the START input of your TCSPC device (i.e. Start measuring time!)
  2. The LED should be "seen" by the single photon detector << 0.1 of the time. This is essential. Single photon detectors have dead times (i.e. when you see a photon, you are blind for 50ns or so). If you always detect the first photon emitted by the LED, you will not be measuring the actual optical pulse of the LED (assuming the optical pulse is shorter than the dead time). Seeing 0.1 or less of the photons allows for the detector to randomly "see" photons anywhere within the optical pulse. This 0.1 requirement is achieved by optically attenuating the light via neutral density filters.
  3. The electrical output of the detector will now be utilised as the STOP input of your TCSPC device. (I.e. stop measuring time. Add the time difference, STOP-START to the histogram).
  4. This process is repeated a million or so times, until a statistically valid histogram is obtained. The histogram of time differences represents the optical pulse of the light source with high temporal resolution.

There are some very good documents and books by Wolfgang Becker of Becker and Hickl (freely available) which describe this process and others in far more detail.

Hope that helps! Rich

  • \$\begingroup\$ @Anindo Ghosh: The average intensity method makes many assumptions, and gives the user no ability to determine the actual rise and fall time independently. There is an exponential decay in the optical output after the current pulse whilst the minority carriers recombine. Hence the optical rise and fall times are far from symmetrical. In addition to this, the technique offers no method to determine the latency, or relative latency between various LEDs or the optical pulses shape. High speed pulse generators are also going to be considerably expensive and modern day TDCs self calibrate. \$\endgroup\$ – Rich Nov 1 '13 at 15:13

Photodiodes are pretty fast. They are used in optical cable communications exceeding several giga bits per second. I've only ever used them at sub giga bits per second but I believe they should be able to measure a period of light that is below 0.1 nano seconds (which is certainly acceptable for most standard LEDs and quite a few laser diodes).

Frame rate footage of an LED being lit and extinguished isn't really required to determine it's ability to pulse light I would have thought.

Measuring time down to periods of pico seconds is something that is beyond my knowledge so maybe someone else will cover this.

  • 1
    \$\begingroup\$ The main problem with trying to measure "latency" of an LED turning on (or whatever) to better than 1 ns accuracy is that every 2 cm of trace or coaxial cable adds about 0.1 ns of delay. And if too many of those "about"s add up, you've got a substantial measurement error. \$\endgroup\$ – The Photon Oct 30 '13 at 0:13
  • \$\begingroup\$ @ThePhoton Actually, the transfer function would be a constant time period added per unit length, while not significantly shaping the pulse profile. Regardless of the delay from emitter to sensor, the shape of the pulse will remain essentially the same. Thus rise time, fall time, and slopes can still be measured accurately regardless of transmission distance, until inverse square fall-off or light contamination kills the light or makes the SNR too poor to sense. \$\endgroup\$ – Anindo Ghosh Oct 30 '13 at 7:10
  • \$\begingroup\$ @AnindoGhosh, 1. The question asked about latency, not rise or fall time. (Of course its possible the OP is confusing these concepts). 2. Once you get below 1 ns you'd better be careful about your assumption that a transmission line is just a delay element with no effect on the pulse profile. \$\endgroup\$ – The Photon Oct 30 '13 at 15:47

They use photodiodes and other high speed optical detectors. You can get photodiodes with very wide bandwidths. This one has a 40 GHz bandwidth, so it should have a rise time of around 9 picoseconds or so (BW is approx. 0.35 / rise time).



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.