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

E-mail: Jack.Smith@cliftonlaboratories.com
 

Table of Contents
Introduction
Test_Setup
Frequency_Measurements_with_3586Bs_Internal_Counter
Improved_Frequency_Accuracy_with_HP3586B_and_3562A
Extracting_a_15625_KHz_IF_Sample_for_High_Accuracy_Off-the-Air_Frequency_Measurement

Revision History
Originally written 24 September 2008
27 September 2008 Fixed grammatical errors
27 September 2008 Added photo of 3568B; added CHU 14.670 MHz plot from K8AQC.
26 May 2009. Revised error in BFO discussion. Added section on extracting 15625 Hz IF signal for high accuracy off the air measurements.

Did you ever wonder what a shortwave receiver designed and manufactured by Hewlett Packard would be like? 

HP actually designed and manufactured several radio receiver models, although they were usually not called "receivers." I have two HP receivers.

One is a vacuum tube super-regenerative receiver, model 417A, covering the range 10 - 500 MHz, with a 5 µV sensitivity. Officially, it's a "VHF Detector" to be used with slotted lines and bridges, such as the 803A. (I have an 803A to go along with the 417A, so all I need to complete a state-of-the-art circa 1955 VHF/UHF impedance measuring facility is an HP608D or E signal generator.)

The other is a synthesized receiver, covering the range 50 Hz (yes, that's 50 Hz, not 50 KHz) to 32.5 MHz, with three bandwidths, 20 Hz, 400 Hz and 3.1 KHz, USB and LSB mode. It's a model 3586B "selective voltmeter."

HP made the 3586B and its 3586A and 3586C variants for the  telephone industry, to measure analog carrier systems, carried over coaxial cable, twisted pair, open wire and microwave radio systems, not to mention LF and HF  radio systems. It's little known in the amateur radio community, but the Bell System had single sideband carrier systems operating not long after WW I ended.

The 3586B is the Bell System variant of the 3586, with the "A" model conforming to CCITT international standards and the "C" model being more an "industry standard" design. If you have the opportunity to acquire a 3586, I would look for the "B" version and also look for Options 003 and 004. Option 004 adds a high stability oven time base (1 part in 10⁷ stability per year) and 003 substitutes a highly desirable 3.1 KHz widest bandwith for the B's standard 2.0 KHz filter. 2.0 KHz is a bit narrow for my liking for SSB and 3.1 KHz is more usable.

The 3586B comes equipped with Western Electric standard slide-in 75 ohm connectors for antenna input. You can purchase an adapter to BNC, pictured at the right, from Pomona Electronics (Model 6519 BNC To Miniature Weco Adapter) for about $30 from Mouser or Newark. Or, you can modify the front panel to accept a BNC connector.

I'm not going to go into the 3586B's features other than to note a couple of the more interesting ones.

As in keeping with the purpose of  a selective voltmeter, the 3586B measures the amplitude of the signal to which it is tuned, with an accuracy of ±0.25 dBm under the right circumstances.

It also measures the carrier frequency with a built-in counter, with a rated accuracy of ±1 Hz, although as we'll see later on this page, it's quite a bit better than this specification.

In keeping with the Bell System's standards, the 3586B centers the filters around 1850 Hz. Depending on which operational  mode you select, you may have to enter an 1850 Hz offset. Other operational modes allow you to enter the theoretical carrier frequency of an SSB signal and be correctly  tuned.

The 3586 does not have AGC as we understand it in the amateur radio community. It's primarily a level and frequency measuring box, and in order to maximize the level accuracy, it is usually used in a mode with 10 dB amplitude window. If the  received signal goes out of this 10 db window, the 3586 switches in or out fixed gain steps to restore the signal to the 10 dB window, which can be disconcerting to listen to, compared with a normal SSB receiver. It's possible to operate the 3586 in fixed gain mode, similar to a normal receiver with the AGC turned off and the audio level controlled via the RF Gain control. Even in the best circumstances, it's not the receiver of choice for quality audio; rather it's a receiver with a rather specialized—but highly useful— toolkit. (One useful feature is a built-in tracking signal generator.)

The last point is that the 3586B has a broadband front end—selective voltmeters were designed to be used on systems with many signals all of about the same peak power, although of high total power summed across all the signals. Hence, it has no front end pre-selection or filtering to speak of. Accordingly, if you live in the shadow of a 50 KW AM broadcast station, you may need additional RF input filtering.

If you wish to learn more about the 3586B, Agilent has put up PDF copies of the 3586 operating and maintenance manuals for free download. Go to Agilent's main page, http://www.home.agilent.com/ and enter 3586 in the search box. You will then be  taken to a page with links to the 3586A, B and C literature. There's actually one set of documents for the A, B and C models, as the small differences amongst the three did not justify separate manuals.

Test Setup

The diagram below shows the test setup I've been tinkering with. In order to maximize frequency accuracy, I've used a GPS disciplined 10 MHz crystal oscillator, a Trimble Thunderbolt, as the master time base. The Thunderbolt has a quoted short term accuracy of 1 part in 10^-9 improving to 1 part in 10^-11 when averaged over long periods. Since the work covered on this page is a matter of seconds, not hours or days averaging, 1 part in 10^-9 is the most appropriate accuracy figure.

Frequency Measurements with 3586B's Internal Counter

The simplest way to measure an off-the-air carrier frequency is to select Carrier mode, enter the frequency on the 3586B's front panel control keypad and then press the Counter button. The right hand display shows the frequency with 0.1 Hz  resolution. (The left display shows the signal level, usually most useful when displaying dBm.) The 3586B's circuitry measures the strongest signal in the passband, and the default 20 Hz bandpass filter is useful to ensure only the desired signal is measured.

I wrote a program in Liberty Basic  to step  the 3586B through a list of frequencies and measure the signal level and frequency error from the nominal frequency entered on the list. In one hour of collection, the software made 26 individual frequency measurements on 17 AM broadcast stations in the Washington DC area and the standard time and frequency stations CHU at 7335 KHz and WWV at 10.000 MHz, with the results below. This data set was taken between 6:30 and 7:30 PM on Tuesday, 23 September, before the local AM broadcast stations shifted to nighttime operation and before CHU faded out or became obscured by shortwave broadcast interference as normally happens in the evening.

I've also found that my 3586B's counter function seems to be biased upward 0.1 Hz, compared against the more accurate techniques discussed later on this page. For a more accurate estimate of frequency error, I would subtract 0.1 Hz from the mean error column.

In a 26 May 2009 E-mail, Burt, K6OQK, notes that all 3586's he is familiar with read 0.1 Hz high and has an interesting observation on this error:

From what I've been able to determine regarding the instruments tendency to ready 0.1 Hz high is that the phone companies would rather see that than 1,849,999.9.  That takes some interpretation! 
 

Rather than collect and display the frequency  reading, my program saves  the error between the nominal frequency and that measured by the 3586B. The FCC's rules, by the way, require AM broadcast band stations to maintain their frequency within ±20 Hz of the assigned frequency. All are well within that tolerance, except for WEMD in the nearby Virginia city of Manassas, which is close to +21 Hz. (I noticed this problem several days ago and sent an E-mail to the station at that time, but have not had a response. Frankly, I don't expect one as budget cutting in the broadcast industry has left little money for technical things.)

For those not familiar with standard deviation, a brief non-technical explanation is in order. Suppose we made 20 frequency measurements of two stations. Station A measured 0.0 Hz error from the assigned frequency in each of the 20 measurements. Its average error is thus 0.0 Hz. Station B measured +10 Hz on the first 10 measurements and -10 Hz on the next 10 measurements. The average, or mean, frequency error of Station B is also 0.0 Hz.

However, the mean obviously does not provide the full picture in this example. The standard deviation is a measure of how far the individual measurements varied from the mean. In this simple example, station A's data has a standard deviation of 0.0 Hz, whilst Station B's data has a standard deviation of 10.0 Hz. (For the mathematically inclined, this is the standard deviation of the entire population of 20 measurements. If the 20 measurements are regarded as a sample of a larger set of possible measurements, the estimated standard deviation is a bit higher, 10.26.)

The mean and standard deviation together give us a good view of how close the average is to the correct value and how scattered the individual measurements are with respect to the mean.

With this information, for example, we note that WEMD is 20.8 Hz high on average, with a standard deviation of 0.12 Hz, which tells us that its transmitter is stable, but off the assigned frequency.

Looking at data for CHU and WWV, two stations that we know are transmitted as close to the assigned frequency as one might possibly expect, we see that CHU's mean error is quite small, -0.005 Hz, or 6.8x10^-10. This is quite a remarkable number, and is likely to be a statistical fluke, as Doppler shift due to propagation can be expected to cause much greater errors. The National Research Council Canada, CHU's owner, makes the following accuracy statement:

Normally CHU's emission times are accurate to 10^-4 s, with carrier frequency accuracy of 5x10^-12, compared to NRC's primary clocks, which are usually within 10 microseconds and 1x10^-13 compared to UTC.

WWV data shows an order of magnitude greater mean error, at 0.072 Hz or 7.2 x 10^-9. The National Instutute of Science and Technology says WWV's frequencies are accurate to "a few parts in 10¹³."

The standard deviation of these 26 measurements is 0.0911 Hz (CHU) and 0.0458 (WWV). This means that the WWV measurements are more consistent, which fits with my experience. The path from Ottawa, ON to Clifton VA at 7335 KHz fades more often, deeper and faster than the 10 MHz path between Boulder CO and Clifton. I make this observation without direct evidence that I can point to (although I hope to collect it soon) but based on experience listening to both CHU and WWV.

The data demonstrates, however, that the 3586B's counter function is far better than HP's quoted ±1 Hz and is more like ±0.1 Hz. (There's an inherent ±1 digit in the last place, which matches the observed ±0.1 Hz counter accuracy.)

It's possible to improve the measurement resolution considerably by measuring the audio beat note output from the 3586B. This can be done with a frequency counter, or a calibrated computer sound card. I used an HP 3562A Dynamic Signal Analyzer, which is a 0-100 KHz spectrum analyzer based on Fast Fourier Transform principles. The HP 3562A's time base is also linked to the 10 MHz GPS-disciplined oscillator, so both parts of the system have a common, high accuracy reference oscillator source.

As I mentioned earlier, the 3586B's center reference is 1850 Hz, so errors from 1850 Hz correspond to errors from the 3586B's set frequency. If the beat note is 1850.1 KHz,  then the received carrier frequency is 0.1 Hz high. (This assumes the 3586B is set for USB mode, of course.)

The HP3562A plots below show WWV at 10 MHz and CHU at 7335 KHz with a 10 Hz horizontal scale. Each division is thus 1 Hz.

Looking at the tip of the waveforms, it's easy to see why the signal as received has error. The Doppler shift spreads or broadens the signal width to, perhaps, 0.1 Hz in the CHU sample and perhaps a bit more in the WWV sample.

Why is the signal subject to Doppler shift? These are skywave signals and arrive at Clifton VA after being refracted via the ionosphere. The ionosphere's effective height changes from time to time and thus the path length changes. This produces Doppler shift in the received frequency, in the same way that a police radar system receives a slightly frequency shifted signal returned from a speeding vehicle. In the case of HF radio signals, of course, the shift is much smaller.

For slow (non-relativistic) speeds, the Doppler shift is:

where f is the frequency in Hz, v is the speed in meters/sec and c is the speed of light, 3x10⁸ meters/sec.

Solving for v, we find:

v = cΔf/f were Δf is the Doppler shift in Hz.

For 0.1 Hz Doppler shift at 10 MHz, we find the ionosphere is moving at about 3 meter/sec, or about 7 miles/hour if I've converted correctly. This is a plausible number, certainly.

The two plots below show CHU at 7335 KHz and WWV at 10 MHz, as their carrier is broadened by Doppler. The horizontal axis is 10 Hz on these plots, so even a small frequency shift is quite visible.

It demonstrates, without question however, how clean a stable ground wave signal can be. (I've been told by WMAL's chief engineer that its transmitter is phase locked to a GPS master clock. When I looked at WMAL's frequency using the high accuracy techniques discussed later, I can find no measurable frequency error within the limits of my equipment.)
 

This appears to be a discrepancy somewhere in my setup or test gear. The plot below shows the 10 MHz master reference. Since this has no propagation issues, being connected by coaxial cable directly from the master oscillator to my 3586B, and since this master reference is the clock signal for both the 3586B and 3562A, there should be no offset error, and the frequency should read 1.850 000 KHz. Instead, it shows 1.849 913 KHz, or 87 milliHz low.

(Added 26 May 2009) When I wrote this paragraph, I was under the impression that all the oscillators in  the 3586B were phase locked to the frequency reference. This is incorrect, in that the BFO is not synthesized but rather is derived from a quartz crystal oscillator. Or, to be more accurate, the BFO is derived from one of several crystal oscillators. The 3586B, for example, has five different BFO frequencies, depending on USB or LSB, and certain instrument functions and options. I won't go into the details here, but the nuts and bolts of it are covered at 8-C-18 of the 3586's maintenance manual. Unfortunately, the BFO oscillators do not have trimmers  to net them exactly on frequency.

An improved method of extracting an accurate frequency sample is to use the 15625 KHz IF, as discussed later on this page.

To better illustrate the spreading seen over skywave propagation, I collected data on WMAL ground wave, CHU at 7335 KHz and WWV at 10, 15 and 20 MHz in spectrogram format.  Note the difference in spread between WMAL's ground wave signal and CHU at 7335 KHz. Energy is spread over several tenths of a Hz in CHU's case. WWV at 10 and 20 MHz, during the time I looked at the signals, happened to be quite stable with little to no fading. WWV at 15 MHz was weaker than either 10 or 20 MHz and had noticeable fading, but not as much as CHU.

The first is WMAL. As the image shows, no Doppler shift is seen, and none is expected since propagation is  via ground wave. The plot below has a total span of 2 Hz, so Doppler will show up rather well.

The weak signal at the right side of the plot is a co-channel station, some 65 dB below WMAL's signal at my Clifton VA location.

Note the offset frequency read, 1.849 907 5 KHz, -92.5 milliHz from the theoretical center. (This is likely due to an error in the BFO crystal.)

The carrier tone measured is 1.849 905 KHz, -95 milliHz from the theoretical center.

The measured tone frequency is 1.849 922 5 KHz, or -77.5 milliHz from the expected value.

The carrier tone is 1.849 9075 KHz, or -92.5 milliHz from the expected value.

The rightmost column should be understood to mean the error from the nominal carrier frequency. CHU, for example, is measured at 2.5 milliHz below 7335 KHz, or 7334.999 997 5 KHz.

I should note that with the 2 Hz span setting, the HP 3562A's spectral bin width is 3.75 milliHz, which places a floor on frequency resolution at this span. Accordingly, the 2.5 milliHz errors in the table above are within the bin  resolution margin of error.

The table below shows the HP 3562A's primary parameter settings.

All the data presented above is based on measuring the 3586's demodulated audio out. Since the accuracy of the audio output is a function of the free running BFO crystal there is room for improvement.

Up to the point of the product detector/BFO, the 3586B's frequency conversion process is locked to the instrument's master time base, or the external GPS or Rubidium or other standard. At the suggestion of Burt, K6OQK, I modified my 3586B to extract a sample of the 15625 KHz 2nd IF output. The 2nd IF output is derived from locked oscillators and hence represents a high accuracy off-the-air signal sample.

This approach is not my original work, but it may be helpful to document how I went about it.

The first question is where to obtain the sample. One branch of the 3586's 15625 KHz IF chain has a limiter, which provides a constant amplitude output for the IF sample. (The limiter output is used in the 3586's frequency measuring circuitry.) The limiter output is, of course, a square wave and of rather high amplitude. A  two section RC filter following the limiter output will restore more of a sine wave waveform, and also reduces the level to a value suitable for general measurement purposes. The series resistor also provides some safety should the output be inadvertently shorted.

Board A22 holds the limiter and the red "X" in the schematic shows the sample point,  the output of U4a. Click on the schematic image for a larger image.