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

E-mail: Jack.Smith@cliftonlaboratories.com

Introduction

I've long wanted a pair of identical RF signal generators for intermodulation tests and other uses. Although I have a collection of signal generators, I did not have two identical models. Recently I picked up two HP 8657A synthesized generators, covering the range 100 KHz - 1040 MHz. The 8657A devices are late 1980's vintage (one has a 1989 serial number and other a 1992 serial) and are controllable from either the front panel or over a GPIB interface.

The only difference between the two generators I have is that the 1992 one has the high precision time base. HP's 1996 catalog characterizes the 8657A as an "economy priced" generator, with a list price of $9,760, plus an additional $1,080 for the high stability oscillator option. 

A major problem with synthesized signal generators is phase noise. I knew that the 8657A generators will exhibit phase noise; the question is how much and how bad or good is it compared with other generators in my shop. KO4BB has a comparison page at http://www.eds-fl.com/Test_Equipment/Signal_Generators.html where he compares the HP 8656A, HP 8657B and HP 3586A. The 8657B makes some improvements over the 8657A in phase noise and frequency resolution and adds pulse modulation input as well as increasing the maximum output frequency to 2060 MHz.

Test Setup

Perhaps the most popular phase noise measurement method used by home experimenters is the software developed by John Miles, KE5FX, as part of his GPIB Toolkit, available for free at http://www.thegleam.com/ke5fx/gpib/readme.htm. Unfortunately, my Advantest R3463 spectrum analyzer is not supported in his phase noise program.

I've instead used a simpler approach; use the Softrock Lite 6.2 receiver in the configuration discussed at Using Softrock as a Panadapter for the K2 as a phase noise measuring instrument. The signal generator under test is offset 50 KHz from the Softrock's center frequency (4898 KHz approximately) and the resulting 50 KHz output signal is fed into an HP 3562A Dynamic Signal Analyzer. The 3562A is a dedicated low frequency (0...100 KHz) spectrum analyzer and tracking generator using digital sampling and FFT  techniques. For serious, calibrated work, I prefer it to using the E-MU 0202 sound card and spectrographic software as it has known calibration.

The data collected with this simplistic approach, of course, assumes  that the Softrock Lite's crystal oscillator and digital dividers have much less phase noise than the instrument under test. I don't know for sure  this is the case, but it seems to make sense and the data is consistent with this being true.

I've also used a notch filter approach in looking at phase noise. For more details see my Oscillator Noise Measurements page. The notch filter measurements concentrate more on wideband noise, whilst the measurements on this page are close in to the oscillator. I've also looked at phase noise from canned crystal oscillator modules at my Canned Osc Phase Noise page.

For each generator, I ran tests with a resolution bandwidth of 10, 2.5 and 1 KHz. This resulted in 15 plots.

The plot below overlays all five generators at 10 KHz span. The horizontal line represents the test setup noise floor with a 50 ohm termination on the Softrock Lite's input and thus is the test setup's noise floor. Note the vertical scale on all plots is 12.5 dB/div, so that 100 db range can be presented.

8657A (no. 1) on top of 8640B

Composite

Composite

There's no question that the HP8640B wins, hands down, for lowest phase noise. Somewhat surprising is the difference between the two 8657A generators. The newest generator, with a 10 MHz precision time  base, has 60 Hz related spurious signals about 10 dB worse  than the older generator. Both, however, meet HP's specifications, although the newer generator is close to the edge of HP's performance specification.

Panasonic's VP8191A does quite well in terms of close in discrete spurious signals for a synthesized generator. However, the VP8191A has quite a bit greater broadband noise than either HP8657A generators.

The Boonton 102D isn't a bad performer, although it has more discrete spurs than the other analog generator tested,  HP's 8640B.

Manufacturer's Specifications

HP provides the following typical performance specification for the 8657A.

I also looked at the harmonic output of these five generators, over the range 0-20 MHz and 0-100 MHz.

The 8640B is a 256-512 MHz cavity-tuned oscillator with digital divider chains for lower frequencies. Each digital divider is followed by an octave low pass filter, which, when combined with the inherent odd-order harmonic suppression of a square wave, makes the output quite clean.

Both the HP8657A and Panasonic VP8191A synthesized generators run the synthesizers at a frequency in the 100 MHz range, down converting with a mixer and fixed oscillator for lower frequency bands, such as the one examined here. Boonton's 102D analog oscillator uses a similar design approach, with a permeability tuned oscillator in the 60-120 MHz range, which is down converted for lower frequencies.

Hence, every generator tested, save for the 8640B, has the potential for non-harmonic related spurious signals on the lower frequency bands.

The data presented below consists of a scan from 0...100 MHz, followed by a scan from 0...20 MHz. The signal source is set to 4937 KHz at -33 dBm in all cases.

A similar approach to exploring phase noise can be applied to the Elecraft K2 transceiver in receive mode using the test setup shown below.

The K2's IF output signal will reflect the composite phase noise of the HP8657A precision crystal oscillator, the K2's internal phase locked loop and variable oscillator and the Softrock's crystal oscillator. We will assume, without a great deal of justification, that the predominant phase noise source in  this setup will be the K2's local oscillator chain.

Sheet 1: Synthesizer

The K2 uses a PLL (phase-locked-loop) synthesizer IC (U4) in conjunction

with a wide-range, band-switched VCO (Q18). The synthesizer provides

approximately +7 dBm output from 6 to 24 MHz, which is then injected at the

transmit and receive mixers (sheet 2). Phase noise performance of the

synthesizer is very good despite its low parts count and absence of shielding.

The reference oscillator for the PLL IC is temperature-compensated by the

components on the thermistor PC board. This circuit works by applying a

variable offset voltage to varactor diodes D16 and D17 to compensate for

drift As temperature increases, the uncompensated oscillator would drift

down in frequency. The thermistor causes a slight increase in the bias voltage

to these diodes as the temperature increases. The relative values of RA-RD

and the thermistor, Rt, (see below) set the rate of gain change with

temperature.

 

U4 provides coarse tuning (5 kHz steps). Fine steps are achieved using a 12-

bit DAC (U5) to tune a voltage-controlled crystal oscillator (Q19), which is

the PLL reference oscillator. The reference oscillator range needed on each

band varies in proportion to the VCO output frequency. To cover exactly 5

kHz in 10 Hz steps on each band, an automatic calibration routine is provided

in firmware. The DAC is swept from its highest output voltage down, and the

DAC word needed to select each 100 Hz step is recorded in EEPROM on a

per-band basis. 10 Hz steps are then interpolated based on the 100 Hz table

data. Crystal X1 in the PLL reference oscillator can be tuned by varactor

diodes D16 and D17 over a range of about 10 kHz, which is required in order

to tune the full 5 kHz on the lowest band (160 m), but still provides better

than 10 Hz resolution on the highest bands.

 

The synthesizer design is unique in that three inexpensive DPDT latching

relays are used to select one of eight VCO ranges, thus requiring only a single

high-Q VCO inductor (T5). The relays are optimally interconnected to allow

for maximum coverage of the nine HF bands, plus a large out-of-band tuning

range. Computer simulation was used to find a relay topology that allowed for

the use of standard 5% fixed capacitors along with the smallest practical

varactor diode capacitance. As a result, the VCO exhibits low noise on all

bands and has a low max/min tuning ratio on each band.

Of course, the 8640B was designed as a general coverage signal generator more than 30 years ago, so this comparison is perhaps unfair or meaningless. Still, it's interesting to note how clean a special purpose synthesizer can be.
 

The HP3562A's dynamic range is limited to about 80 dB, which places a limit on the noise level we can observe. We can, with a small degree of extra effort, extend this range by 40 or 50 dB by discarding part of the less interesting spectrum, i.e., the carrier. If we can reduce the 50 KHz carrier of our test signal by, say, 40 dB, we have, in essence shifted our 3562A's dynamic range downward 40 dB, providing the functional equivalent of 120 dB range.

Where would one find a 50 KHz notch filter? One possibility is to build one. Or, if one has an classical audio distortion analyzer, it's possible to use the built-in notch. The classical (by that I mean before the days of digital signal processing) audio distortion analyzer worked on a simple principle. Set the signal to be analyzed at a pre-defined level. Then use a very narrow notch filter to remove the signal. Measure the residue and call it distortion and noise.

Most distortion analyzers have a post-notch output port, so running the test signal into a distortion analyzer and looking at the post-notch output with the 3562A provides exactly what we are looking for—an output with the carrier notched. I happen to have an HP333A distortion analyzer, so I decided to give it a  try.

Rather than using the Softrock and its internal crystal oscillator, I decided to use an double balanced diode mixer and a 10 MHz crystal signal source. I tried two 10 MHz sources, one being precision GPS-corrected oscillator out of a Trimble Thunderbolt and the second being the high stability time base output of an HP8657A signal generator.

HP's 333A, by the way, uses the same Wein bridge phase shift circuit as found in its classic audio oscillators, such as the HP200CD. I've written about the HP200CD, viewable by clicking here.
 

As good as the 333A's notch is, it's not infinitely narrow, as seen in the image below extracted from the operating and service manual. This means our notch technique works best when looking at noise more than a few percent from the center frequency. Since our center frequency is 50 KHz, we should not place too much emphasis on the data within, say, 5 KHz of the 50 KHz center frequency as the notch attenuation changes rapidly over this range.

With the 40 dB notch and 30 dB gain, the topmost graticule line represents 40 dB below the applied carrier level. The 8640B's phase noise at, say, 25 KHz from the carrier frequency, is thus nearly 120 dB below the carrier level, or -120 dBc. If you compare the plot below with the earlier data the improvement in dynamic  range is obvious. Once you move more than 5 KHz from the carrier, by the way, there is little difference in phase noise between the two HP8657A signal generators, with the noise plots being within a few dB of each other.

The signals at 100 KHz are the 2nd harmonic of the  50 KHz carrier.