ben-wpan/cntr

Arbitrary-precision frequency counter
=====================================

Theory of operation
-------------------

The arbitrary-precision frequency counter counts clock cycles of a
frequency source that is assumed to be free from drift. It compares the
count with the host's system clock. If the system clock is synchronized
with an accurate NTP reference, measurements with arbitrarily high
accuracy can be obtained.

In practice, this is limited by the the frequency source's drift and
the time one is willing to wait. If NTP maintains the system time
with an accuracy of +/- 100 ms, obtaining measurements with an
accuracy of +/- 1 ppm would take about 28 hours.

Additional error sources, such as the round-trip time when requesting
a sample from the microcontroller, are also considered in the accuracy
calculation.

The counter consists of a board based on a C8051F320 microcontroller
and the control software on the host. The microcontroller counts
events in a free-running 16 bit counter that is regularly read and
extended to 32 bits. The 32 bit counter is periodically queried by
the host.

The microcontroller's counter can count at a frequency of up to 3 MHz.
(SYSCLK/4)

In order to protect against transmission errors not detected by USB's
CRC, which are occur relatively often, each packet is protected by a
CRC-32 and an inverted copy of the payload. Corrupted packets are
rejected by the host.

The 32 bit counter wraps around at most once very 21.8 ms. The 32 bit
counter wraps around at most every 1431 s. The host extends the 32 bit
counter to 64 bits, and calculates frequency and accuracy from the
count and the run time of the measurement application.


Performing a measurement
------------------------

To perform a measurement, connect the CNTR board's probe input to the
clock source and then run the "cntr" application on the host. An
accuracy goal (in ppm) can be specified on the command line (see
below).

The host polls the microcontroller every 100 ms and displays the run
time (in seconds), the measured frequency, and the accuracy achieved
so far.

Measurements can be stopped by pressing ^C or by specifying an
accuracy goal. At the end, the total number of events counted and
communication statistics are displayed.


Updating the firmware
---------------------

The protocol revision and the build date of the firmware of the CNTR
board can be queried with "cntr -i".

To update the firmware, run
cntr -r; sleep 1 && dfu-util -d 0x20b7:0xcb72 -D cntr.bin


Known issues (version 2 hardware)
---------------------------------

- the input circuit only works up to about 1 MHz. The problem is that
  we discharge too slowly though the base of Q1, which in turn keeps
  the transistor turned on too long.

  An alternative design that sets R2 to zero, adds a 47 Ohm termination
  resistor in parallel to VR4, and places a 1 kOhm resistor between
  VR4 and Q1 works up to about 2 MHz, but accepts a lot of HF noise
  and is very sensitive to the signal amplitude.

  Some test results with a ~1.8 m RG-174 cable, square wave bursts
  with a 50% duty cycle and ~ 5 ns raise/fall time:

  Design       Frequency  Source amplitude Input amplitude
               (nominal)  (nominal)        (measured)
  ------------ ---------- ---------------- ---------------
  Original     1 MHz        700 mV         ~700 mV
  Alternative  2 MHz      1.6 V            ~800 mV
               1 MHz      1.5 V            ~750 mV

  The test consisted of setting the frequency and adjusting the nominal
  source voltage in increments of 100 mV for the lowest voltage at
  which which ten consecutive bursts of 50000 cycles each were all
  received correctly.

  The source has an output impedance of 50 Ohm, so voltage at the probe
  input (indicated in the table) is roughly half the nominal source
  voltage in the alternative design.

- the MMCX connector is hard to solder because of its large thermal
  capacitance and surface
- the lateral pads of the MMCX connector could be wider
- manufacturing tolerances can push the USB connector a bit inside
  the board, making its signal pins overshoot the pads
- the ground fill under the LED causes a secondary cathode contact
  (which is harmless on this case)