[time-nuts] HP 3586A/B/C entirely referenced to 10MHz:

Bert, VE2ZAZ ve2zaz at yahoo.ca
Tue Mar 22 07:48:01 EDT 2011

Hi Bill,

It is a great analysis you did. The truth is I found the AD5932 more or less 
randomly while searching for DDS chips. I computed the accuracy with a 100KHz 
MCLK and found that I would get better than 0.001 Hz. I knew that I would never 
be bang on, so I did not try to pinch more bits. This of course later became 0.3 
Hz as I increased to a 10MHz MCLK. But the unit was already on the bench, 
installed on the A22. If I had to do it again, I would consider other DDS chips, 
as long as the price is similar.

In the end, since I know what the offset is, I just take it into account in my 
frequency calculations. Even with a better DDS, I would have still taken it into 
account. Spectrum Lab can go quite far in its averaging calculations. Besides, 
my main objective was to stabilize the HP 3586 more than try to be exactly on 
frequency. So in that sense, I have met my objectives.

I am glad to see that my SLM mod will be used as a starting basis for future 

Thanks for your input.

Bert, VE2ZAZ


Message: 6
Date: Mon, 21 Mar 2011 18:32:07 -0700
From: WB6BNQ <wb6bnq at cox.net>
To: Discussion of precise time and frequency measurement
    <time-nuts at febo.com>
Subject: Re: [time-nuts] HP 3586A/B/C entirely referenced to 10MHz:
Message-ID: <4D87FC17.892DE535 at cox.net>
Content-Type: text/plain; charset=iso-8859-1


I am curious to know what caused you and Paul to select the AD5932 device ?

Admittedly, I haven't verified the Analog Devices simulator with real 
but I suspect their simulator is spot on or damn close.  Using the Adsim page I
looked at a few different DDS?s to see what could be done.  With little 
cost better choices are available allowing better on-frequency accuracy relative 
the offset values of the AD5932.

The problem with AD5932 is the frequency tuning word [FTW] is too small.  So,
clearly, increasing the FTW would give an immediate improvement as to accuracy.  
simple low pass filter would clean up the spurs as they are all associated with 
clock frequency and well removed from the fundamental signal.  Some DDS 
included an uncommitted internal comparator stage (notably the 9834 and the 
that would serve well for squaring the signal after filtering.

I ran simulations for two different DDS devices.  I picked ones that operated 
off of
5 volts of which there is damn few good ones.  The first one is the AD9834 with 
a 28
bit tuning word with a 10 MHz clock.  Here are the results:

13775 = 13775.0059366226 Hz = error of +0.0059366226
14125 = 14124.9969601631 Hz = error of -0.0030398369
14275 = 14275.0144004822 Hz = error of +0.0144004822
16425 = 1642500.01311302 Hz = error of -0.0088095665
16625 = 1662500.01639128 Hz = error of +0.0020265579
16975 = 1697500.01281500 Hz = error of -0.0069499016
17125 = 1712500.00596046 Hz = error of +0.0104904175
17475 = 1747500.00238419 Hz = error of +0.0015139580

As you can see, with the additional 4 bit tuning word, the error improves for 
except 17125 where it is equal.  The second run was upping the frequency by 100
times to reduce the size of the filter components.  For the AD9834, this did not
turn out well at all.  The wave form had a hard staircase appearance due to the 
clock rate relationship (5:1) to the higher output frequency.  The same problem
exists for the AD9851.  So, I scrapped that whole idea.

The second run was using the AD9851 with a 32 bit tuning word with a 10 MHz 
Here are the results:

13775 = 13774.9989517033 Hz = error of -0.0010482967
14125 = 14124.9992884696 Hz = error of -0.0007115304
14275 = 14275.0004306436 Hz = error of +0.0004306436
16425 = 16425.0005036592 Hz = error of +0.0005036592
16625 = 16624.9996982515 Hz = error of -0.0003017485
16975 = 16975.0000350177 Hz = error of +0.0000350177
17125 = 17124.9988488853 Hz = error of -0.0011511147
17475 = 17474.9991856515 Hz = error of -0.0008143485

As you can plainly see, increasing the tuning word by, yet, another 4 bits 
for shifting the error further to the right.  Maybe enough to put it beyond the
resolution of the total measurement system and thus, perhaps, removing it from 
systemic error list (i.e., less to worry about in the calculation).

Unfortunately, Analog Devices has stopped producing some of their easier to use 
bit DDS devices.  The current run of 48 bit DDS?s are way more complicated and
specialized, have issues with the clocking methods (time nut unfriendly), 
using lower voltages, they are harder for the hobbyist to mount to a board and 
are more expensive.  Truly a sad circumstance for the occasional hobbyist.


More information about the time-nuts mailing list