Flea_small.jpg  The PFM Flea  pictures/flea/Flea_small.jpg

<< back to page 1                                                                page 2


Jitter_diagram.gifA clock signal, like the one that is used in a CD player, is generated by an oscillator circuit. This is a piece of electronics that contains an amplifier, which is brought to oscillate by positive feedback. By inserting a frequency-dependent element in the feedback-loop (a crystal or a filter network), a narrow-banded signal is obtained. In the ideal world, this signal would contain only one single frequency, which on top of that would also be very stable. But due to the hard fact that, unfortunately, the electronics are not ideal, this signal consists of multiple frequencies (harmonics and various intermodulation products) and has a certain bandwidth. Also, the desired frequency is not entirely stable. Factors like (thermal) noise and non-linear behaviour of the components cause the signal to be slightly modulated in frequency.

Because of this, the flanks of the clock signal do not correspond exactly with the right time-interval (UI). This error in time of a periodic signal, in this case the square wave of the clock signal, is called jitter. The order of magnitude (in the case of audio) is in the pico-seconds (ps) range. If the jitter is caused completely by random events, the distribution of the time error will be a normal or Gauss distribution, as is the case with many other natural processes. The flank then deviates an equal amount in the positive as in the negative direction. The difference between these two extremes is the peak-peak value of the jitter. The RMS-value equals one standard deviation σ (= sigma) of the distribution.

Pierce_osc.gifOn the left it is shown how the oscillator is often constructed in most digital audio-equipment. With only a few discrete parts and an inverter, the clock signal is generated. The inverter is configured as an amplifier by means of Rf, and it is usually integrated in one of the other IC's that are present in the equipment, like the DAC, the decoder or the digital filter. But this way, it obtains it's power from the same supply-pins, or through the same voltage regulator as the rest of the circuit. This causes high-frequency noise signals from other parts of the IC to have easy access to the oscillator through the common supply-lines, where they can cause jitter. The other way around, the harmonics generated by the oscillator can easily reach other parts of the IC, or other parts of the circuit. This situation is far from ideal. By fitting the oscillator externally and feeding it with a low-noise voltage regulator and a separate power supply-module, the cross-talk is greatly reduced and jitter is reduced. From this point of view, The Flea is developed.


Measuring jitter in the pico-second range is not a problem nowadays, with a suitable (expensive) analyzer. Unfortunately, the possibilities in this case are limited by the available equipment: an Agilent Infiniium oscilloscope model 54835A and a LeCroy Waverunner oscilloscope, model 104MXi. Because their functionality is limited regarding jitter, it is not possible to filter the jitter-signal. The scope will therefore display the entire jitter-spectrum, while the area of interest for audio is actually only the low-frequency portion. The results will be higher because of this, and that's why this measurement is only a relative comparison.

Two measurements were done on The Flea, while fitted in the ezDAC. The first scope-image shows the histogram of the 16.9344MHz clock signal in a standard unmodified CD57. Here can be seen that the jitter of the clock signal is 66.9ps RMS. The second image shows the 24.576MHz clock signal of The Flea as this is present in the ezDAC. Because of the different clock frequencies, the comparison is not entirely valid, but it can be clearly seen that the jitter in the second image is much smaller: 8.9ps RMS for The Flea. The third measurement with the Waverunner scope shows slightly higher results: 13.8ps RMS jitter was measured here.

Standard 16.9344MHz clock
in a CD57
24.576MHz clocksignal of The Flea
in the ezDAC
24.576MHz clocksignal of The Flea
in the ezDAC

Fitted in a Marantz CD67mkII-OSE

Fitted in a Marantz SA8400

In the SA8400 the digital logic runs on a 3.3V power supply instead of 5V. For this player, the applied clock signal must be adapted to this to prevent overdriving of the inputs. As the XO-module is specified for a supply voltage of 3.3...5V, this can be easily accomplished by feeding the XO-module with 3.3V as well. The output voltage of the Flea can be adjusted by altering the two resistors of the voltage divider. Besides the resistors, the output voltage is also determined by the forward voltage drop (Vf) of the green LED. With the formula given below, the output voltage can be calculated:

Vout = Vf ( R6 + R7
     or:       R6 = R7 ( Vout  -1)
The green LED that's used here has a Vf of 1.94V. For a 3.3V output voltage, R7 is chosen 470Ω for example (originally 820Ω for 5V) and R6 then becomes 330Ω (originally 1k3 for 5V). By measuring the exact voltage across the LED in practice, the voltage divider can be adjusted if needed.

Fitted in a Rotel RCD-991

5V Flea, 16,9344MHz. The clock signal enters at pin 3 of IC206 (74HC04).
Remove C211, C212, R202 and X201. R201 can remain. Connect the clock output of The Flea to the empty pads of C211.

More will follow soon...

Fitted in a Music Hall CD25.2

> Thanks to Everett for providing these pictures!

  • DSC03241_cr_small.jpg  Overview with the extra power supply on the top-left
  • DSC03243_cr_small.jpg  Close-up of the Flea PCB

Fitted in The ezDAC

In the ezDAC, the digital logic also runs on 3.3V. So this Flea is adapted to that voltage as well. This time, for R7 the original 820Ω is used, and for R6 two 1k2 resistors are used in parallel, of which the second one is mounted on the bottom side of the PCB. This makes 600Ω for R6 and together with 820Ω for R7, this gives a perfect 3.3V output voltage.

> Big thanks to 'Mags' for supplying this Flea PCB!

<< back to page 1