Clifton Laboratories 7236 Clifton Road  Clifton VA 20124 tel: (703) 830 0368 fax: (703) 830 0711

E-mail: Jack.Smith@cliftonlaboratories.com
 

Revision History
17 August 2009 Original

Table of Contents
Phase_Noise
Short_and_Medium_Term_Instability-Frequency_Counter_Data
Short_and_Medium_Term_Instability_-_Receiver_Test
What_is_the_Si500_Specification
 

Looked at on the gross level, there's no obvious difference between the Si570, the Si500 and a simple low performance crystal oscillator. The spectrum analyzer image below shows the outputs of these three devices, all at 7996 KHz, captured with an Advantest R3463 spectrum analyzer. The four discrete spurious signals seen, in fact, all originate inside the R3463.

This is, of course, a very undemanding test of an oscillator, as the spectrum analyzer's bandwidth is 10 KHz in the sweep and the R3463's internal phase noise is not all that great. It would take an extraordinarily poor oscillator to show problems with this test. Its purpose is to see if there are significant discrete spurious signals detectable and none are seen, other than ones internal to the R3463.

The reason I asked Tom to supply the Si500 at 7996 KHz, however, is so that I could use a narrow band crystal notch filter to reduce the carrier by approximately 100 dB. (At the deepest part of the notch, the carrier is down approximately 115 dB. This part of the notch is quite narrow, however, and the Si500 was not quite centered in frequency to reach the deepest part of the notch.) This allows much greater amplification to be used in the spectrum analyzer and effectively extends its dynamic range. I've used a crystal notch filter to look at phase noise before at http://www.cliftonlaboratories.com/oscillator_noise_measurements.htm which has some data at 13.500 MHz taken with an earlier version of the notch filter as well as data at 7996 KHz with an improved version of the filter. Data on the 7996 KHz notch filter and the test protocol is contained at http://www.cliftonlaboratories.com/CannedOscNoise.htm towards the bottom of the page. To improve the spectrum analyzer's noise figure, I added a Mini-Circuits ZFL-500LN 30 dB gain low noise preamplifier between the notch filter output and the spectrum analyzer input.

Still, it's reasonably clear that the difference between the Si500 and Si570 is not all that much, although the Si570 looks to be better in the important close in range up to 10 KHz from the carrier.

Before we get too carried away with how good these two "synthesizers in an IC" are, note that even a mediocre crystal oscillator is hugely superior. It has noise just a few dB above the pre-amplifier/spectrum analyzer combination noise floor and is 30 dB or more better than either the Si500 or Si570. (The crystal oscillator is a Clapp design, along the lines developed by G3UUR for measuring crystal motional parameters. The 7996 KHz crystal in the example is an inexpensive microprocessor crystal, as are  the ones in the notch filter.)

So far, so good with the Si500. Not so with the next test.

One important criteria of an oscillator useful in amateur radio receivers is stability, short, medium and long term. I suppose I should identify these periods. The easiest is long term stability; how much does the oscillator drift when observed over a period of several hours. By medium term stability, the observation period is in a matter of a few minutes at most and short term is a second or so.

Long term stability is perhaps the least problem as a general matter, After all, for most of the history of amateur radio, we had free running LC oscillators with drift measured in several KHz during the first hour or two of operation, although some equipment did much better, with a few hundred Hz drift from warmup to thermal stability, which became necessary as SSB replaced AM voice communications. Regardless, long term drift requires at most an occasional touch  to the tuning control.

Intermediate term drift is more  troublesome, as, at the worst, it  requires the operator to keep one hand on the tuning control all the time.

Short term instability is even worse, as it occurs too fast to be compensated for with the tuning control. Rather it manifests itself as a "burble" or warble on received signals.

Attempting to measure the intermediate and long term stability of the Si500 with a frequency counter quickly showed something odd with the Si500. The plot below shows the measured frequency of the Si500 and Si570. (Disregard the two jumps when I disconnected the Si500 and connected the Si570.)

Note the "fuzz" on the Si570's frequency reading. Each main vertical division is 50 Hz in this plot and the fuzz is approximately 5 Hz. This can be regarded as short term instability. In addition, note the jumps and shifts of the "average fuzz" which is intermediate term instability. The Si570, in contrast, is nearly blameless in this regard, showing a clean, stable frequency history with just a small bit of warm up drift.

I should also note that using a frequency counter understates short term jitter, as will be seen in the later analysis. This is because the counter averages out the number of cycles measured during the gate.

After spending a day or so chasing down possible explanations for the frequency counter "fuzz" I ruled out ground loops and other possible explanations. The final test used a 9V battery to power the Si500 and isolated the RF output with a 1:1 isolation transformer. No change. Combinations of attenuators, high impedance or low impedance counter input, adjusted triggering levels, and the like all made no difference.

The next step was to listen to the Si500 on a receiver, in this case a Racal RA6790/GM. I set the Racal for USB mode, and tuned it to provide approximately a 1 KHz beat note against the Si500 output. To avoid ground loops, I powered  the Si500 with a 9V battery during this test, and the connection to the Racal receiver was via stray pickup, with a short wire in  the Si500's output jack radiating a signal to the Racal's receiving antenna. (Tom's PCB includes a 3.3V regulator and may be safely powered from a 9V source.)

After a few seconds listening, the problem was obvious. The Si500 has an objectionable short term burble. A quick comparison with  the Si570 with identical power and connection showed a clean signal.

I've recorded short WAV files of both oscillators. Click on the buttons below to listen.

I also ran the Racal's audio out into a desktop PC running ARGO spectrogram software with the results below.

The difference is astounding. The Si500 has about a 5 Hz peak-to-peak random FM modulation with short term duration.

In addition, there's an intermediate term instability of around 10-15 Hz.

Both of these are objectionable and the audio excerpts reveal why I regard the Si500 as not usable for amateur radio transmitters or receivers.

The excerpt below from the Si500 data sheet provides the typical and worst-case jitter.
 

This leads me to wonder whether the Si500's short term crud is this jitter and that accordingly the Si570 is operating within its specification. The Si570's phase jitter spec (depends on frequency and type of Si570) is on the order of 0.3 ps. Plugging this into an 8 MHz fundamental, the corresponding peak-to-peak frequency jitter is 38 Hz. In fact both the Si570 and Si500 are doing quite a bit better than their specifications, but the Si500 is not good enough for serious receiver use.

Tom has learned that the Si500 uses a 4 GHz LC tank as the frequency determining element, whilst I believe the Si570 has a quartz crystal time base. This likely explains the difference in short and intermediate term stability.

I should add, of course, that the Si500 (and the Si570, for that matter) was not developed for amateur radio receiver or transmitter uses. Rather, it's intended to be a time base used in digital circuitry. A change in frequency of the magnitude I observed is insignificant for all but a few digital application. On the other hand, we expect our receivers and transmitters to exhibit a very high degree of short and medium term stability. Hence, it's not surprising that the Si500 comes up short in our rather demanding application.