Agilent 33622A Waveform Generator - Review

Table of contents

RoadTest: Keysight 33622A Waveform Generator

Author: Instructorman

Creation date:

Evaluation Type: Independent Products

Did you receive all parts the manufacturer stated would be included in the package?: True

What other parts do you consider comparable to this product?: Agilent 33522B 30 MHz Waveform Generator and Tektronix AFG3102 100 MHz Arbitrary/Function Generator.

What were the biggest problems encountered?: A high pitch squeal emanates from the generator all the time when it is powered down in standby mode. On a couple of occasions the generator locked up during boot up and had to be power cycled.

Detailed Review:

Review of the new Agilent 33622A Waveform Generator

April 26, 2014


Note to readers:  This review has an introductory part that provides background content on waveform generators in general. The introductory section is intended to inform those who are new to electronic test equipment.  To get straight to the meat of the review, scroll down to the Detailed Review header.


Introductory background on waveform generators

Electrons are the core stuff of electronics. Fundamental electronic concepts like current and voltage are characteristics we attribute to electrons.  As important as electrons are to the essence of electronics, humans cannot see them.  In spite of there invisibility, there is plenty of circumstantial evidence they are real.  Anyone who has forced too much current through a resistor can attest to the heat, light and smoke electrons can liberate.  Humans have developed sophisticated methods for manipulating electrons and pushing them around for our collective benefit.  To facilitate the development of these sophisticated methods we invented instruments that provide a proxy sense of what all the many invisible electrons are doing in our circuits.

Every student of electronics engineering at some point in their studies connects abstract theory to practical reality by exploring the unseen realm of electrons using laboratory instruments that allow us to manipulate and observe circuit behavior.  Instruments like oscilloscopes let us visualize the changes in voltage over time within a circuit.  Waveform generators let us create time varying voltages that stimulate the operation of circuits and simulate information, the essential commodity and very reason for the existence of most electronic circuits.


A function generator, or waveform generator, is a staple test instrument found in most college and university labs, on many electronic engineers’ benches and in the homes of many electronic hobbyists.  The function generator allows the user to specify the parameters of several standard waveforms, most often sine waves, square waves, ramps and pulses.  Each of these standard waveforms serves a purpose in evaluating and troubleshooting electronic systems.  Standard function generators produce a single output wave shape and allow control over amplitude and frequency. Higher performance generators may have a second output, a wider selection of wave shapes, ability to sweep between two frequencies, ability to modulate the wave shape, adjust the phase and synchronize to an external trigger signal.  Higher performance generators also tend to achieve higher signal frequencies and offer much better accuracy and stability.  Test instrument technology is in a period of rapid innovation that is allowing companies to add new capabilities with each new product cycle.


Contemporary function generators provide today’s skilled user with several useful enhancements not available in older generators.  Local instrument set up and control is now a breeze due to the addition of menu driven graphical user interfaces.  Remote instrument control is a commonly included feature, enabled through USB, GPIB, or TCP/IP interfaces.  The most innovative and useful enhancement, in my opinion, is the additional functionality offered by arbitrary waveform generation.


Arbitrary waveform generation allows the user to specify, in exquisite detail, the exact shape of the generated voltage waveform.  Although sine waves, ramps, and pulses are “real” signals and they are great for performing well-understood laboratory tests of system performance, they fall short of representing the messy signals encountered by electronic systems disbursed in field applications. Sensors buffeted by the fluctuations of their environments, mundane or exotic, generate real signals.  Custom electronics generate real signals to perform particular functions.  Noise infests real signals, real signals have missing pulses, runt pulses, reflections, or unique shapes that defy pushbutton selection.  To fully characterize and troubleshoot real systems it is best to inject real signals into them.  However, obtaining a real signal in the lab can be problematic.  If the signal necessary for testing comes from an exotic environment like the inside of a nuclear reactor, the vacuum of space, or the bottom of an ocean, it may be impractical to generate the signal in the electronics lab from the original source.  If the signal is proprietary, or dangerous, generating it from the original source may be out of the question.


Recently I worked on a project where it would have been illegal for me to generate the original signal under investigation in a lab by myself.  Suffice to say this project involved testing prohibited weapons.  On another project, the signals generated by the sensors were complex, low amplitude and almost drowned in background noise.  On these projects, having a good arbitrary waveform generator in the lab addressed environmental, legal and safety barriers that in turn allowed research to progress effectively.  On projects like these, a sampling oscilloscope captures a high fidelity record of the original signal.  Software (like Excel) is used  to manipulate the record to generate derivative signals that simulate faults or stresses like added noise, transients, missing segments, delays, reflections, et cetera. I do not regret my decision to obtain a high quality arbitrary waveform generator for our lab.

Detailed Review

I'll start by sharing the results of my basic comparative performance tests.  Later in the review I will describe some user experience issues I dealt with while performing the performance tests.

The Agilent 33622A reviewed here is an R&D grade waveform generator with arbitrary waveform capabilities.  The 33622A is a member of the 33600A series of generators, which in turn appear to be an extension of the Agilent 33500B series of generators.  Built with the same chassis and front panel as the 33500B series, the 33600A series offers higher maximum frequency, more standard waveform memory (4 mega-points v. 1 mega-points) and a different suite of user installable upgrades.  Owners of generators in the 33600A series can purchase vouchers to enable additional capabilities as needs emerge.  For example, an 80 MHz 33611A can be upgraded to 120 MHz bandwidth by purchasing a 336BW1U Bandwidth Upgrade ($696 Canadian dollars at time of writing).  A list of all options and upgrades available from Agilent is located here.


In 2012, Element 14 offered the Agilent 33522B Waveform Generator for review.  The 33622A and 33522B are so similar in appearance that you have to look carefully at the printed model number next to the Agilent logo to tell they are in fact different products.  To get a really solid sense of both generators, I recommend reading the four 33522B reviews written by Element 14 Road Testers located here.

The 33522B is the top end offering in the 33500B series.  For this Road Test, I am reviewing an Agilent 33622A, the top end offering in the 33600A series.  For several years, I have used a Tektronix AFG3102 Arbitrary/Function generator.  The table below compares selected specifications for all three generators.  Specs in the table show the capabilities of the instruments on my bench.  Both Agilent generators were provided with additional arbitrary waveform memory enabled. 



(all are 2-channel)

Maximum sine frequency


Maximum pulse frequency


Minimum pulse width (ns)

Maximum amplitude

(Sine into 50 Ω)

Standard arbitrary sample memory

Resolution (bits)

Maximum sample rate

Agilent 33622A



Max 4 Vpp


≤4Vpp to 120 MHz

≤10Vpp to 50 MHz


(optional 64 MSa/channel)


1 GSa/s

(250 MSa/s with filter OFF)

Agilent 33522B




10 Vpp

(20Vpp open circuit)

1 MSa/channel

(optional 16 MSa/channel)


250 MSa/s

Tektronix AFG3102





(20Vpp open circuit)

128 kSa/channel


1 GS/s (max 16k points)

250 MSa/s if >16k points


For this review, I set up each of the three generators to produce identical signals, then captured oscilloscope traces from each device to allow comparison of performance.

Running acceptance tests

I referred to procedures in the Windows CHM Agilent TrueForm Series Operating and Service Guide when performing the following tests.  As it happens, I have some of the test equipment recommended in the service guide and was able to perform some of the quick verification tests.

The test equipment I had available included an Agilent 53230A 350 MHz Universal Frequency Counter/Timer with the ultra high stability oven controlled crystal oscillator option (OCXO), an Agilent 34461A 6½ digit multimeter and a Tektronix MDO4104-3 Mixed Domain Oscilloscope with 1 GHz analog bandwidth and a 3 GHz spectrum analyzer.

Internal Timebase Verification

All instruments used in the timebase verification test were warmed up for over one hour before testing began.  Ambient temperature, measured with a two wire thermistor probe on the Agilent 34461A, varied between 23°C and 24°C.  Test frequency in all cases is 10.00000000 MHz












10.000 000 54



10.000 000 54




10.000 000 81



10.000 000 81




9.999 999 8



9.999 999 8



Sine wave spectrum test

The MDO4104-3 spectrum analyzer was set up to display an average of 16 traces.  Automatic markers were enabled to mark up to five peaks with at least 5 dB excursion above a -70.0 dBm threshold.  Reference level on the spectrum analyzer was set to +10.0 dBm with each generator set to output 0 dBm into 50 Ω.  These settings were selected to prevent any over-driving of the spectrum analyzer input.  Over-driving the input would cause clipping and reduce fidelity by adding harmonics to the spectrum. The signal from each generator was coupled to the spectrum analyzer with a 3 foot long RG58C/U cable.

During these spectral purity tests I noted the harmonics of the sine waves generated by all three instruments varied over frequency.  I rationalized this discovery by assuming that several techniques are probably used inside the generators to produce sine waves over such a wide range of frequencies and that each technique has associated distortion characteristics.

In the images below each instrument is generating a 30.000 000 00 MHz sine wave at 0 dBm amplitude into the 50 Ω input load of the spectrum analyzer.  Keep in mind that 30 MHz is the top end frequency for the Agilent 33522B.  The Agilent 33622A has a 120 MHz top end and the Tektronix AFG3102 has a 100 MHz top end.  Also, in the interest of full disclosure and fairness, a calibration error in my AFG3102 was discovered during this review.  The error certainly affects fidelity of pulses, and may be the source of the spurious content around 18 MHz in the third spectral image below.

As I mentioned previously, the harmonic distortion changes with output frequency.  Take a look at the spectrum traces from the AFG3102 and 33622A instruments at maximum specified operating frequency.

These two instruments have considerably less harmonic content at their top end frequencies than at much lower frequencies.  In all cases the harmonic content is well below the fundamental, so there isn't any concern here over quality.  It is just interesting to note the changes in harmonic content vs. output frequency.

Comparative jitter tests

I made some measurements that allowed relative comparison of jitter performance between each of the three generators on pulse edge jitter.  To make these measurements the oscilloscope was set up for 50 Ω input on channel 1.  The acquisition mode was set to Sample, the waveform display set to infinite persistence, and the zoom feature was used to find a rising edge following the trigger edge.  Vertical sensitivity was adjusted to produce a full screen displacement with horizontal time base set to 400 ps/div.  The horizontal waveform histogram was activated with a ±3.00 mV vertical range.  All generators were cranking out 20 ns pulses at 25 MHz with tr and tf set to 8.4 ns which is the minimum on the 33522B.  Amplitude was 1.000 Vpp.

The histogram above appears to occupy about one minor division.  With a major division representing 400 ps, a minor division is one-fifth of that or 80 ps, or stated another way, a maximum of ±40 ps of edge jitter .  The published jitter spec is  <1 ps RMS as measured on an Agilent E5052B signal source analyzer in the 10 to 40 MHz band.  My test is within that band, but is not an RMS measurement, rather is is a distribution measurement of jitter over time.  The shape of the histogram suggests the vast majority of edges occur very close to the center of the span, so the RMS jitter value will be lower than ±40 ps.  Another way to get an indication of jitter, if not a definitive measure, is to look at the distribution of measured periods over time.  If there was no jitter, all periods would be identical.  Using an Agilent 53230A counter/timer I produced a histogram of period measurements over more than 275,000 periods.  The results are shown below.

By subtracting XMin from XMax, obtained from the screen image above, the range of periods is found to be 73 ps. This number is comparable to the 80 ps value estimated from the oscilloscope histogram.  The image below shows a 17 ps standard deviation, indicating a fairly tight clustering of periods around the average.

Now, on to the jitter comparison of the other two instruments.

The same jitter test performed on the Agilent 33522B produced the result below.

The results are very similar to those obtained from the Agilent 33622A.  About ±40 ps of peak-to-peak edge jitter.

Jitter on the Tektronix AFG3102 is shown in the screen capture below.

I'm seeing just a slight increase in jitter compared to the two Agilent generators.  Maybe 100 ps peak-to-peak.  The published jitter spec for this generator is 200 ps RMS.  Looks to me like it is doing better than that, at least from the perspective of this test scenario.

Arbitrary waveform generation comparison - Agilent is vindicated

In my review of the 33522B I set up a comparison of arbitrary waveform generation between the 33522B and the AFG3102. Both instruments were generating an arbitrary waveform from identical source files.  The Agilent instrument produced a smoother version of the waveform, which I initially thought was preferable to the Tektronix version which retained quantization steps.  However, as jbridge commented in my 33522B review, "
The output is certainly nice and smooth but perhaps the Tektronix is more closely following the original input arb data".  Good point.  An arbitrary waveform generator should faithfully reproduce the data.  Unless a smoothing filter is selected by the user, the output should look just as it did when it was captured or created in software.

Turns out I did have a filter selected on the Agilent and I only found out that it existed tonight while writing this review.  By way of excuse making I'll say that there are a lot of detailed and difficult to remember features packed into modern test equipment.  I ran the tests again to see if they were reproducible (they are) and in the process I discovered a filter selection feature I had overlooked in the 33522B review.  This filter feature, available on arbitrary waveforms, works great.  It can be switched off, set to normal or to step. It is available on both the 33522B and the 33622A.  Take a look at what is does in the images below.


In the image above, the blue trace shows the output from the AFG3102.  This waveform has steps 100 ns apart, which I expect because it was generated by a data acquisition card running at 10 MHz.  The yellow trace shows the 33622A waveform produced by the same file.  It is beautifully smooth because the arbitrary filter is set to normal.  The image below shows what happens when the 33622A filter is switched off.

Look at that! Now the two generators produce identical outputs that correspond exactly to the description provided in the waveform file.  I really like this feature and I can see where smoothing filters, available in the Agilent generators but not in the AFG3102, could be useful.  Switching the filters on removes the sharp edges in the quantized waveform which reduces high frequency noise.  I made a short video to illustrate the significant difference the filter can make on quantized arbitrary waveforms. Take a look!


How do you choose between two capable instruments?

Okay, so these quick comparison tests prove that all three of the generators are quality instruments with high caliber performance that meet or exceed published specs.  So, if you need a laboratory grade arbitrary function generator for R&D or troubleshooting and you have a budget of under $10k,  how do you choose between the offerings from Agilent and Tektronix?  Some part of the decision will come down to preference for the aesthetics of one instrument over the other.  I prefer, for example, the size of the Tektronix.  The buttons are larger and further apart so I can use my low accuracy fingers to better effect. Commonly accessed functions like Frequency/Period, Amplitude, Offset, Duty, Width, and modulation modes are dedicated to front panel buttons on the Tektronix, so there are fewer menus to navigate than on the Agilent.  The display is larger on the Tektronix as well - good for older eyes, and its larger size means a more complete description of the waveform can be presented.

However, the display on the Agilent is brighter and higher resolution.  The font size can be enlarged on the Agilent if necessary.  Overall, the Agilent display wins my vote.  The real decision making points are more subtle for me.  There are several small unassuming features available in the Agilent generators that pop up from time to time and make me think "ah ha, the engineers at Agilent were thinking of the user when they put this feature in!"  Things like the option to enter amplitude in dBm as well as mVpp, Vpp, mVrms and Vrms, but only when the output load is set to 50 Ω.  Or the ability to set sample rate for arbitrary waveforms.  On the Tektronix you have to figure out the period you need based on the sample rate and the number of samples in the waveform file - a simple calculation to be sure, but one that is not necessary if you can set sample rate directly.  Agilent lets you set phase values in degrees, radians or seconds.  Tektronix insists phase be declared only in degrees.  Then there is that really cool output smoothing filter for arbitrary waveforms.  I wish my Tektronix had that filter.  Memory may be the deal breaker.  The AFG3102 is seriously hobbled by its tiny broom closet size 128 k point waveform memory.  64 M point memory, available in the 33622A, feels like a wide open prairie field in comparison.

Not everything was sunshine and lollipops

While I was generally delighted by the all the useful features and problem solving clout packed into the Agilent 33622A, a few incidents furled my brow and compelled me, like an Olympic figure skating judge, to shave points off the ratings for this instrument.  My first distress came the moment I plugged the 33622A into mains power. 

You may have to crank the volume up to hear the audio in the video clip below.  The microphone frequency response was not great at higher frequencies, but my middle age ears have no problem hearing this.  How would you like to listen to this all day long?

The shrill squeal from the 33622A disappears as soon as current starts flowing through the circuits, but while it is in standby it makes this noise continuously.  It seems to come from the left front side of the chassis and sounds to me like a switching power supply.  The firmware upgrade that mentioned muting the output in power off conditions made no difference.

Also, on two occasions since receiving the unit on April 14th it has locked up during the power on boot up sequence.  It gets past the first monochrome splash screen then the progress bar grinds to a halt about 30% into the second phase of boot up.  Cycling the power is the only way I have found to bring it back to life.  I must say, the Tektronix AFG3102 has never locked up, during boot up or during operation.

When I first attempted to access the 33622A remotely using a LAN, I encountered a Java security problem. My limited Java knowledge meant it took me a awhile to solve the problem, but I did. After upgrading to Java 7 update 55 I found that none of LAN connections that allow instrument control worked anymore (Agilent and Tektronix).  When the icon for instrument control is clicked, the following error message appears:


In spite of the suggestion that this error can be ignored (why else provide an Ignore button?), it can not be ignored.  The web interface will not load until this security issue is resolved. To prevent this error from stopping access to the instrument control page, I found the following procedure effective.


Start the Java Control Panel, select the Security tab and click on Edit Site List.


Add the instruments IP address to the Exception Site List.  Don’t forget to put http:// in front of the address.

You may get a security warning pop-up as shown below.  I clicked on Continue guided by my assumption that Agilent will probably not compromise personal information on my computer.


The next time I clicked on the instrument control button, the application ran without incident,and has worked correctly since.




There was a time when I was not all that fond of Agilent test equipment. That time is in the past.  I am very impressed by my recent review experiences with Agilent instruments.  They are well built, attractive devices that are very well documented with multiple forms of user support.  The performance is really professional grade and if you take the time necessary to get to know them, they are easy to use. 

If I was looking to purchase an arbitrary waveform generator today, I admit I would be leaning toward the Agilent 33600A series.The 33600A series satisfies my current needs for R&D and troubleshooting more effectively than my old Tektronix AFG3102.

April 27, 2014

The 33622A locked up again this morning during boot-up.  On a hunch I disconnected the USB cable from the rear of the unit.  The next power up resulted n a successful boot-up.  The computer was off during the unsuccessful boot-up cycles, so I'm wondering if the lock-ups are related to USB enumerating problems.  Not sure.  This needs further investigation.

All the best,


May 2, 2014 update

Additional jitter tests and Amplitude flatness test

I`d like to thank jbridge and Michael Kellett for commenting on my review and requesting additional tests to answer questions they have about the Agilent 33622A.  I think a benefit of a community based forum like Road Test is the ability to interact with other interested members and help answer their questions.  If I have the ability, time and necessary equipment I am happy to do what I can to answer questions related to equipment I`ve obtained through Road Test.

I'm adding additional jitter tests for each of the three generators taken at 25.974 MHz as requested by jbridge.  The idea is to see how the generators perform when they are generating a pulse with a non-integer multiple of the sampling rate of either 250 MSa/s (40 ns/sample) or 1 GSa/s (10 ns/sample).  My previous tests at 25 MHz did not really stress the generators in that 25 MHz is generated as an integer multiple of either sample rate.  The suggested frequency of 25.974 MHz has a period of 38.5000385 ns.  That value does not slice into an integer number of 40 ns or 10 ns samples.  The hypothesis to test is that this frequency should generate more jitter than found at 25.0 MHz.  Here is what I found:

First, the Tektronix AFG3102.

Yes, the AFG3102 produces more jitter at 25.974 MHz than at 25 MHz.  The amount of jitter at 25.974 MHz appears to be about 120 ps v about 100 ps at 25 MHz.

Next, the Agilent 33522B.

Hard to tell from the `scope capture if there is a difference in histogram width, so I tried the 53230A single period measurement.

I didn`t use the 53230A on the 33522B in the original set of tests, so I have nothing to compare with.

Finally, here is what happened with the Agilent 33622A on the more difficult jitter test.

Once again, it is difficult to detect a difference in the jitter histogram, so I went to the OCXO counter. 

There is a difference.  Standard deviation increased from 17 ps to 25 ps and peak-to-peak deviation increase from 151 ps to 220 ps.  Based on these tests it appears jitter does increase at 25.974 MHz compared with jitter at 25 MHz, however, not by a lot.  These are well behaved instruments.

Next, I ran some sine wave amplitude flatness tests.  To perform these tests I configured the 33622A as follows:


Amplitude500 mV RMS
Sweep start frequency200.000 000Hz
Sweep stop frequency200.000 000 kHz
Sweep time120.000 s
Hold time5.000 s
Return time0.000 s
Trigger sourceManual
Sync sourceChannel 2 (Channel 1 tracking)

This configuration will generate a smooth (linear) sweep over two minutes from 200 Hz to 200 kHz.  Channel 1 was connected to the input of an Agilent 34461A 6½ digit DMM.  Channel 2 was connected to the Agilent 53230A counter and to Channel 1 on a Tektronix MDO4104-3 oscilloscope.  Using the second channel to monitor frequency and waveshape allowed channel 1 to drive the multimeter without the burden of additional parallel loads.  The 34461A was set up as follows:


AC voltsRange auto, Filter >3Hz and Filter >200Hz (see explanation below)
Vertical scaleAuto
MathStatistics - Show

I ran the test twice, once with the DMM AC Filter set to >3Hz, then again with the AC Filter set to >200 Hz.  The noticeable difference is the number of samples taken over the two minutes sweep duration and the associated increase in noise with more samples.  Below are screen captures from the meter and Excel spreadsheet graphs for the sweep showing the effect of >3 Hz and >200 Hz AC filter selection.

On the left is the result of the sweep with the >3Hz filter turned on.  The image on the right is the same sweep with the >200 Hz filter turned on.  Notice the noise appears to be greater in the >200Hz test and that almost 3X as many samples were taken with >200Hz turned on.  Also notice the span is less than 1 mV RMS in both cases.  That seems reasonably flat to me over that range of frequencies.

The same data presented in larger Excel graphs is presented below.

Every test I do provides another perspective on not only the device under test, but on the other instruments used to perform the tests.  I`m enjoying this.


May 8, 2014 update

Ok, I'm going to report on a few more tests I've carried out on the 33622A inspired by an exchange of comments with jbridge (see below).  First, an interesting demonstration of how the 33622A can be coaxed into generating a 500 MHz sine-like wave, then another jitter test with a OCXO high stability source to see if the elusive <1 ps jitter claim can be verified.

Squeezing 500MHz out of a 120 MHz generator

The idea, as suggested by jbridge, is to create a two point arbitrary waveform. One point at max amplitude, the other at min amplitude.  That waveform can then be "played back" or generated at a sample rate selected by the user.  The sample rate determines the frequency generated by the 33622A.  For example, if the two point waveform is sampled at 1 MSa/s (1 μs/sample) the resulting output should be a square wave at 500 kHz (2 μs period).  Let's see if that works.

First, there is a minor operational detail to explain.  The minimum arbitrary waveform accepted by the 33622A, at least from the front panel edit menu, is 32 points long.  This is not a problem.  I simply entered a 32-point waveform composed of alternating max high (+10 V) and max low (-10 V) points.  I noticed an odd anomaly when playing with the front panel arbitrary edit screens.  There seems to be a left over error screen from firmware ported from the Agilent 33522B generators.  The 33522B has a minimum 8-point arbitrary file size and an error will be displayed if the user attempts to save a file smaller than 8 points.  The 33622A has a 32-point minimum, yet when attempting to store a file of say 10-points, the error message below is displayed.

Oops.  Back to the experiment.

The image below shows the edit screen with the 32-point file composed of alternating max high and max low points.

This file was then loaded into channel 2 of the 33622A, as shown below.  Note the output filter is turned off and sample rate is set to 1 MSa/s.

The output waveform captured on an oscilloscope is shown below.  Note the frequency is indeed 500 kHz

The maximum sample rate with the filter off is 250 MSa/s.  Setting sample rate to 250 MSa/s produces a sad looking waveform at 125 MHz - already 5 MHz beyond the 120 MHz spec limit.  See below.


With the output filter set to Normal, the max sample rate jumps to 1 GSa/s.  This should allow us to generate a waveform at 500 MHz.  There is a catch.  The output filter reduces the amplitude of the signal from 20 Vpp down to about 25 mVpp and there is a periodic wobble, but the oscilloscope measures the frequency at about 500 MHz.  See screen capture below.

The waveform shown above is really just a curiosity and not of sufficient quality to be used in most applications.  However, at sample rates below 1 GSa/s, nicer sine waves are produced, but affected by the filter roll-off.  The chart below gives an idea of what amplitudes can be generated at frequencies above the maximum sine wave spec of 120 MHz.


Sample rate (MSa/s)
Frequency (MHz)Amplitude (pp)
300150600 mV
400200409 mV
500250266 mV
600300144 mV
70035070 mV
80040042 mV
90045028 mV

An example of the waveform quality at 300 MHz can be seen below.  The Agilent 53230A counter measured this waveform at 300.0000138 MHz.


So, it is possible to generate sine waves of moderate quality and low amplitude well beyond the 120 MHz design spec for this generator.  See the spectrum image below taken when the 33622A was generating 350 MHz.  There is a lot of harmonic and spurious content in there.

As an aside, the 30 MHz Agilent 33522B was able to generate 125 MHz at 250 mVpp using similar techniques.

Another attempt at measuring jitter on the 33622A

For this experiment I connected the 10 MHz oven controlled reference oscillator on the Agilent 53230A to the reference input on the Agilent 33622A.  This should reduce clock jitter significantly.  When set up to run off an external reference the 33622A shows Ext Lock in the upper right hand corner.  I set the edge times on the pulse to 2.9 ns, the fastest the 33622A is capable of.  The histogram on the oscilloscope is shown below and the 53230A single period statistics are also embedded below.

Under these conditions the peak-to-peak jitter is less than one minor division at 400 ps/major division, so less than 80 ps. 

Standard deviation reads 16 ps with peak-to-peak jitter at 156 ps, about the same as previous measurements.  This is with a 25.974 MHz pulse with 20 ns width, 2.9 ns tr and tf and 1.0 Vpp into high Z load.

Although I made these measurements under consistent instrument conditions I wonder about the validity of the settings.  A change of vertical sensitivity on the oscilloscope changes the edge slope, which in turn changes the histogram width.  I selected a vertical sensitivity of 20 mV/div to produce a steep slope and maintain reasonable linearity.  At lower V/div settings the slope becomes more vertical, but it looks like the waveform is seriously clipping the input amplifier, resulting in distortion and invalid readings.  So, what V/div setting should be used to get the most representative jitter measurement?  I'm not sure.

If any of the other reviewers have the equipment necessary to test jitter, I'd love to see their results.

I won't be able to add anything else to this review for about a month.  Thanks for reading!


  • Yes, a 100MHz scope is going to attenuate any higher frequency components of the output even more than the filters on the 33622. Especially given that you have three or four filters in series (the digital filter in the 33622, the image reject filter in the 33622, the intrinsic bandwidth of the 33622 amplifier which depends on the gain setting being for 10Vpp or 4Vpp and then the input circuitry of your 100MHz scope).


    The jitter measurements are even more problematical as, if the jitter really is less than 1ps, equipment with low enough phase noise or jitter itself to be below this is likely to be very expensive. The 53230A counter that Mark uses is top of the range and pretty expensive if bought new but it still only has 20 psecs resolution and has an intrinsic "jitter" of around 13 psecs rms when measuring a jitter free source such as delay down a cable. But the fact that Mark was able to measure any variation between the 25MHz and 26MHz cases seems to indicate to me that perhaps the generator is contributing more than 1psec (which would be down in the noise).

  • Peter,


    Thanks for the heads up on the Waveform Builder Pro license.  When I attempted to register the serial number on my unit to get the license the web site said thanks very much, then booted me out.  If you don't mind sharing the name of the Agilent agent that helped you, that would be great. Perhaps in a private message.


    My unit came with the 64 M sample memory option, no IQ player or SEC options.



  • I tried through the web site three times with no joy, I finally raised a ticket and it got sorted


    Did you have success registering the 33622A atleast


    another question, did you get any additional add-ons ?, I received additional 64M ram which was nice, did anyone get different extras ? oxo10 pehaps ?

  • Hi Peter,


    I applied for one too, but never got an answer.



  • By the way, have you all applied for your forever licence for wave builder Pro, I had a little trouble due to my generator licence not being accepted but Agilent where wonderfully supportive and I now have a full licence for the software. If you have not and run into difficulties I will pass on the name of the Agilent Agent who helped me so well.


    It is only available for a limited time.

  • Well I would love to experiment beyond the 120Mhz limit of this generator but I am currently limited in what I can observe due to my crappy oscilloscope (100Mhz Siglent), If I am selected for the MDO then that would all change of course but until then I will have to be limited to speculation


    I do speculate that if you create say a three sample sawtooth it may well come through at 300Mhz even with the filter, just at a significantly lower amplitude and perhaps a bit curved... I will just have to dream about it for now lol

  • I don't disagree with what you say, certainly regarding the scope. Also the counter has a resolution which is 20 psecs and in the Agilent demo the rms jitter of the counter is 13 psecs so if the 33622A has jitter of only 1 psec it wouldn't register.


    But, there still remains the puzzle that the measured jitter increased between 25MHz and 26MHz which makes sense for the generator but it is surprising for the counter - though perhaps the timing clock within the counter works better at 25MHz than 26MHz.


    One way perhaps to tell is to do the following:


    1.) Use the OCXO of the counter as an external reference on the generator.

    2.) Use a BNC-T to split the signal from channel A to channel B as well via a meter of cable (as per the Agilent video).

    3.) For both 25MHz and 26MHz do two (or three) statistical one-shot measurements on the counter

    a.) measure the delay from channel A to channel B (the jitter free measurement)

    b.) measure the period on channel A

    c.) possibly also measure the period on channel B


    IF the jitter is all down to the counter then measurements a,b and c will give the same rms values and differences between 25MHz and 26MHz will only arise if there is a change in counter resolution/jitter at different frequencies.


    On the other hand, if the rms of measurements b.) and c.) are significantly higher than that of a.) and the difference increases for 26MHz vs 25MHz then I think the increase can be assigned to the 33622A and not the counter.


    Though I can't claim to be in anyway an expert, I've spent more than 30 years in research and one thing I have learned is that if something is odd then it is worth investigating even if the results appear wrong.


    Anyway, I think this last experiment is worth doing (easy for me to say since you are the one who will need to do it if it is done ) because it will give you information about your counter as well as the generator.

  • I have been doing some research on the web regarding jitter measurement.  My current thinking is that the equipment I have may be limiting my ability to verify Agilent's < 1 ps jitter claim.  The jitter measurement floor is very likely greater than 1 ps on the MDO41014-3 I'm using, although I was unable to locate a published specification in a quick search.  This article from the May 2004 issue of High Frequency Electronics describes a method for arriving at a jitter measurement floor value (or values, as there are several ways to quantify jitter) that I may try at a later time.  I also think Razvan's advice to clock everything from the same source is wise.  I should try triggering the 'scope from the OCXO source in the counter in addition to using the same source as a clock reference for the 33622A.


    At the end of the day I think I'll come to the conclusion that jitter in the Agilent 33622A is probably lower than jitter in any other piece of test equipmet I have and I'll be happy with that.


  • It is interesting that Agilent limit the sample rate when the filters are not applied. I wonder why - perhaps unfiltered high frequency components cause trouble in the output amplifier.


    Mark's results above (which I presume are similar to yours) show that you can get a usable but small signal out at much higher frequencies. For example if you want to test an amplifier you might only want a 50mV input signal anyway.


    The 33622's approach of allowing higher frequencies at lower amplitudes I think may have been a response to companies like Rigol with their DG4162 where they get high headline frequencies by having lower-gain options on the amplifier whilst Agilent previously had a fixed gain (on the 33522) and combined this with attenuators.


    I'm perhaps more intrigued by the jitter measurements. Do you have any kit to measure the jitter on your 33622 to see if it really is less than 1psec?

  • The 33622A will allow a similar thing


    I have had mine already outputting a 125Mhz square wave, the firmware limits the sample rate to 250Mhz without the filters on so assuming two transitions per cycle then that would be 125Mhz


    as soon as you add a filter then you can increase the sampling up to 1Ghz but the filters them selves will kill any usable amplitude if you try to initiate fast swings in output, it is easy to try, create a waveform of 32 samples (The min) and simply make it a square wave, it will output at 125Mhz easily but the minute you apply the filter it almost disappears in amplitude, there is still a signal but not even close to what you want


    with some experimentation it could be possible to go way beyond the 120Mhz but not with a predicable amplitude