Introduction
For the first part of my review, I aim to carry out verification tests on the AFG31052, carry out the same tests on a Rigol DG1022, Uni-T UTG962 and the AWG element of the MDO34 that I also have in my possession. For some of the tests, the Picoscope 3404A USB Oscilloscope with builtin waveform generator was added into the list. Both the MDO34 and Pico 3404A are only single channel devices and are sometime omitted from a test or have reduced data. The tests themselves are based around the verification tests from the service and calibration manual for the AFG3000 series, freely available to download from the Tektronix website. I have included a few other tests that were of interest to me.
The majority of the measurements are made using the MDO34, a 53220A53220A counter from Keysight and a Fluke 8846A 6.5 digit bench multimeter. All of these instruments have an external UKAS calibration certificate, so the readings should be within specification for those instruments. All the instruments were left for one hour prior to taking measurements to ensure that they have stabilised.
Frequency verification
This test is directly from the Tektronix manual. As the clock is shared between both output channels, the test is only conducted on Channel 1 and only at one frequency of 1 MHz. The measurement is made for both a sine and a square wave. The results can be seen below, the expected readings are from the specifications for the AFG31052.
| {gallery} Frequency measurement results |
|---|
Frequency readings for AFG31052, DG1022 and UTG962 |
Frequency readings for MDO34 and 3404A |
Summary of statistical data for frequency tests |
Output comparison plots for frequency, variance and standard deviation |
Output comparison plots for outliers and span |
Apart from the 3404A, all the instruments are extremely close to the nominal 1 MHz output. Both the Tektronix instruments have the lowest number of outliers that fall outside of the 6 standard deviation limits. The MDO34 does have the highest frequency span though. The other three instruments have the majority of their data outside the statistical limits of the AFG31052, indicating that their output is not as stable as the Tektronix instrument.
Converting this data into deviation from the nominal 1 MHz value, in ppm format gives the following table.
It is clear to see that the 3404A cannot meet up to the specifications of the AFG31052, but all of the other generators are well within the deviation tolerance. The data can be plotted for better visualisation with the 3404A omitted from the plot to avoid minimising the rest of the data. Both Tektronix instruments are seen to have a similar deviation, followed by the DG1022 and finally the UTG962 at around 0.9 ppm deviation.
As stated previously, the instruments are left to warm-up before taking the measurements. For the AFG31052, I collected readings just after switch on and then after warm-up to see what the difference would be. A summary of the data can be seen below. The main point is that the spread of the frequency measurements is reduced after the instrument has warmed up. The overall number of outliers, falling outside the 6 standard deviation limits, had improved by just under 5%, although this is seen as a reduction in the upper band as the lower band saw more outliers after warm-up.
This lower spread of frequency data can be seen in the histogram plots above. Both plotted to the same X axis scale, the histogram on the right is seen to be narrower. Both histograms follow a standard bell curve distribution.
To aid with instrument warm-up, the AFG31052 has a builtin timer that can be activated from the utility menu. When activated, a 20 minute countdown timer is displayed along with the option to abort the countdown if desired.
Phase displacement
Phase displacement is not actually covered in the Tektronix verification tests, but I was curious about the phase relationship between the two output channels. For the first test the outputs were left in phase as the frequency was adjusted, for the second, the phase was adjusted with the channels locked and held at 1 MHz and the final test the phase was adjusted on one channel only whilst the frequency was again held at 1 MHz. Being only one channel devices, the MDO34 and Pico 3404A were omitted from this test.
| {gallery} Phase Displacement Test Results |
|---|
Results of phase displacement between channels with frequency adjusted |
Results of phase displacement locked between channels at 1 MHz |
Results of phase adjustment between channels at 1 MHz |
The initial test showed a huge difference between the DG1022 and the other two waveform generators that were much more stable. The phase angle between the two channels did drift more as the frequency was increased. At 50 MHz sine wave, the AFG31052 showed 3.5 times the drift from 0 degrees displacement than the UTG962 showed.
With the channels phase displacement locked and held at a 1 MHz output, the AFG31052 showed a very stable output as the phase of each channel was increased in parallel. The differential across the adjustment range was substantially lower than the DG1022, but was agin beaten by the UTG962.
The final test shows comparable results across all three instruments. The AFG31052 shows just as good stability as the UTG962 and whilst the DG1022 was stable across the test range, the actual value was close to 2 degrees higher than the nominal.
An extension of the phase measurements was to monitor the variance over 10000 samples at 1 MHz and the channels locked at 0 degrees.
| {gallery} Phase displacement histograms |
|---|
Histogram plot for AFG31052 |
Histogram plot for DG1022 |
Histogram plot for UTG962 |
The histogram plots show some interesting results. The UTG962 is a near prefect bell curve. The 3 x standard deviation side bands are symmetrical around the nominal 0 degrees value. The predominant displacement at +0.05 degrees. The AFG31052 is a close second, it shows a slight lean towards a positive variance from the 0 degrees with the predominant phase displacement at 0.20 degrees. Bringing up the rear again, is the DG1022 with a clean lean towards a more positive predominance and also showing a much wider spread. The rep dominant displacement reading was at 0.10 degrees.
The table and plot below summaries the statistical data.
Whilst the data shows the admirable performance of the AFG31052, it is trumped on occasion by the UTG962. The AFG31052 does have a slightly tighter span from minimum to maximum and hence a lower standard deviation. It does have a much higher average than the UTG962 and is actually close to the average displayed by the old DG1022. With each waveform generator compared to the 3 x standard deviation side bands produced from their own data sets, the AFG31052 shows 3.91% of outliers (i.e. those values falling outside of the side bands), the DG1022 shows a similar 3.6% of outliers. Incredibly the UTG962 shows only 0.07% of values outside its own side bands.
Comparing the other two waveform generators to the side band limits established by the AFG31052 data set, shows that the DG1022 is no match for it with a total of 19.52% of outliers. Yet again though, the UTG962 outperforms the AFG31052 showing only 0.7% of its phase displacement values fall outside of the tighter side bands for the AFG31052.
DC offset
The DC offset is carried out in a similar manner to the AC amplitude tests, but the output is set to a DC voltage. The DG1022 does not actually have this facility, therefore the measurement was made with a pulse output with a high duty cycle. The three test points are taken from the Tektronix procedure for the AFG31052. Channel 2 of the DG1022, also has a 2.50 V limit to its DC offset capability, the Pico 3404A is limited to 2.00 V and the MDO34 to 1.25 V.
| {gallery} DC Offset Test Results |
|---|
DC Offset results for AFG31052 |
DC Offset results for DG1022 |
DC Offset results for UTG962 |
DC Offset results for MDO34 |
DC Offset results for 3404A |
The table below summarises the test results in the form of percentage tolerance deviation.
The plot above provides a better visual comparison. Channel 2 of the DG1022 was the worst performer and Channel 1 wasn't that much better. The MDO34 and the 3404A were a bit of a mixed bag, but did perform better than the Rigol unit. The UTG962 did produce some good results, but could not match the AFG31052 that operated had the least deviation from the nominal values for both channels.
AC amplitude
This test is the measurement of the peak to peak voltage of the output. As the amplifier for each channel is different, the tests must be carried out on each channel of the instrument. A set of test points is specified within the Tektronix calibration procedure for the AFG31052. The tests are conducted with a 50 Ohm pass through terminator installed directly on the channel output to provide a load. To improve accuracy of the tests, the resistance of the terminator is first measured to establish a correction factor for the tolerance bands. The output is measured using the bench multimeter, so the test frequency is kept at 1 kHz to keep it with the capability of the multimeter.
| {gallery} AC Amplitude Test Results |
|---|
Amplitude readings for AFG31052 |
Amplitude readings for DG1022 |
Amplitude readings for UTG962 |
Amplitude readings for MDO34 |
Amplitude readings for 3404A / 5244B |
A graphical representation of the test results can be seen below.
Instead of utilising the tolerance limits for each individual instrument, the tolerance limits calculated for the AFG31052 were applied to all of the instruments. The output differential is converted to a percentage to allow for a complete comparison. Both the MDO34 and Pico units have their output limited to 1.5 Vrms and channel 2 of the DG1022 is limited to 2.0 Vrms by design.
All of the instruments fall within the tolerance band for the AFG31052 output. Averaging out all of the results, shows that the MDO34 has the highest differential, quickly followed by one of the channels of the UTG962. Channel 1 of the AFG31052 has one of the tightest tolerances, but the second channel cannot match this. This pattern is seen with the other two channel instruments, with one channel showing a better tolerance than the other.
AC Flatness Test
This test is also prescribed why Tektronix and measures the AC amplitude linearity at higher frequencies than previously measured. This test requires a reference level to be verified using a bench multimeter at 1kHz and the remaining measurements carried out with an RF Power meter. As with the other amplitude test, a 50 Ohm through terminator is used to provide a load to the waveform generator and both channels are tested.
The Fluke 8846A is accurate up to 300kHz for AC voltage measurement, so can be used for the initial reference level. After this, an RF power meter should be utilised for the higher frequency measurements. To meet this requirement, I invested in a small RF power meter courtesy of eBay. I am also interested in an RF meter for measuring the power output of the insulation testers I have with wireless communications, so this is a worthwhile investment for me - and being eBay it didn't break my piggy bank.
A similar procedure is followed as per the AC Amplitude Measurements with the impedance of the through terminator verified to produce a calibration factor to be applied to the tolerance values. The test setup is quite small and the RF power meter can be powered from the USB port on either the AFG31052 or the 8846A.
The results of the measurements can be seen below for both channels of the AFG31052. The reference at 1 kHz come out at 3.98 dBm, the other two measurements with the 8846A are comparable at 3.95 dBm for 100 kHz and 3.97 dBm for 300 kHz, which is the maximum frequency input for the 8846A.
At 1 MHz, the measurement is swapped out to the RF Power Meter, but the value drops to 1.10 dBm for Channel 1 and 1.40 dBm for Channel 2, clearly out of tolerance values. The readings only go into tolerance at 5 MHz and then stay in tolerance for the rest of the readings. Given that this is happening on both channels, I am currently putting this down to the performance of the RF Power Meter. Although it can read down to 1 MHz, it may be that its accuracy suffers.
To verify this, I carried out the same flatness test on one channel of the DG1022 and UTG962 and saw comparable results. It is unlikely that I have 3 instruments from different manufacturers and different ages, all with the same fault. Note that for the DG1022 the 25 MHz measurement was actually made at 20 MHz.
To overcome the issue, I have made another eBay investment in an AE20401 from Ascel Electronics. This is a kit version, so I have a little work to do before I can repeat these tests. However, it has a power meter specification much more suited to this application, DC - 500 MHz bandwidth and -55 dBm to + 30dBm, means I can compare measurements between the 8846A and the RF power meter and I do not need an attenuator.
My aim will be to build up this instrument and repeat the tests on the AFG31052 along with the rest of the function generators.
AM Modulation Depth
To test AM modulation, I followed an application note from Siglent for measuring the depth and carrying out a verification calibration.
Siglent - Measuring Modulation Index
The test signal is a 1.0 MHz, 1 Vpp sine wave, with a 10 kHz AM modulated signal. The modulation depth is set to different points to test throughout the range. The measurement was then made with the spectrum analyser on the MDO34, set for a relatively low bandwidth around the 1.0 MHz centre frequency. The number of averages is increased up to 128 to improve accuracy. Only the AFG31052, DG1022 and UTG962 can output a modulated waveform and the DG1022 only has the function on Channel 1, unlike the AFG31052 and UTG962 that can output modulated waveforms on both channels. As an extension to the test method from Siglent, I tested across five different modulation depths.
On the display of the AFG31052, the carrier and modulated waveform are displayed as two different waveforms, which initially look quite crushed, but is rectified by changing the waveform size on the screen. As the different parameters are selected for adjustment, the AFG31052 highlights these on the waveform. It was found that InstaView would not work on this function. The UTG962 displays the output as a combined waveform, the changes basic shape as the output is adjusted. The DG1022 does not have e a graphical display in the same manner as the other two units.
| {gallery} AWG Screen Displays |
|---|
AFG31052 Modulated Waveform Display Screen |
AFG31052 Frequency Selected |
AFG31052 AM Frequency Selected |
UTG962 at 20% modulation depth |
UTG962 at 100% modulation depth |
The spectrum analyser screens were captured for Channel 1 of each of the instruments.
| {gallery} Modulated Waveform Spectrum Analyser Screens |
|---|
AFG31052 Spectrum Analyser Screen at 80% Modulation Depth |
DG1022 Spectrum Analyser Screen at 80% Modulation Depth |
UTG962 Spectrum Analyser Screen at 80% Modulation Depth |
Visually, the output of each of the functions generators was similar. The spectrum analyser picked up the frequency spikes without any problems, enabling the modulation depth calculations to be made. The following table details the measurement results and calculations.
Across the test range, both the AFG31052 and the UTG962 performed similar. AT 100% depth, the UTG962 slightly out performed the AFG31052. The poor DG1022 showed its age, with an increasing inaccuracy as the modulation depth was increased.
It was also noted that the centre frequency dBm level was slightly more erratic on the DG1022 in comparison t the other two units.
Square Wave Duty Cycle
This is one of the two measurements applied to a square wave output. The square wave only has one duty cycle setting of 50%, this can obviously be varied by switching to a pulse output. There is no requirement to measure this from Tektronix, it just falls into place with the rise/fall time measurements that are a requirement. The 53220A53220A counter I have can measure duty cycle as well as using the MDO34 and I always find it intriguing to compare different measurement methods.
I carried out the duty cycle measurement using the same test procedure for the rise/fall times specified by Tektronix. This calls for a 10 MHz waveform at 1 Vpp into a 50 Ohm terminator and then a second test at 10 Vpp into a 10 dB attenuator. Not all of the instruments can provide this output, so for the DG1022, the test frequency is 5 MHz and for the 3404A it is 1MHz. Voltage wise, the second channel on the DG1022 can only generate 3 Vpp, the MDO34 can only generate 2.5 Vpp and the 3404A only 2 App.
The AFG31052 square wave output at 1 Vpp can be seen below.
The measurements from both sets of instruments are tabulated below and show good correlation.
The maximum deviation from any of the waveform generators was 0.706%, measured on the 53220A53220A for channel 2 of the AFG31052 which a 10 dB attenuator fitted. Curiously both the channels showed a slightly higher duty cycle with the attenuator in circuit, unlike the majority of the other waveform generators that showed a slight drop.
Changing the output attenuation also provides an opportunity to take an initial look at the InstaView system. The objective of the InstaView subsystem is to provide an alternative visual appreciation of the waveform being generated without the use of a separate oscilloscope to measure the output. Prior to using the InstaView, the cable propagation needs to be entered by either running an automated test or manually entering via the keypad.
With the propagation delay entered, the desired channel can be activated. To further enhance InstaView, it is possible to rearrange the screen so that only the one channel in use is displayed, the waveform can then be enlarged by sweeping the plot upwards so that it covers around 3/4 of the screen. For some reason, sweeping up the screen to this height was a bit hit and miss and the graph spent most of the time covering half the screen, that was much more consistent to arrange.
Once activated, the InstaView screen offers a peak to peak and positive and negative voltage values along with a timescale across the X axis. I could not find any cursors to enable user defined values. I did find discrepancies between the peak to peak voltage values displayed by InstaView and those displayed on the measurements from the MDO34 for some of the tests.
With the scope set to a 1 MOhm termination, the AFG31052 displayed a peak to peak voltage of 1.967 V and the scope displayed 1.960 V.
With a 50 Ohm termination set on the oscilloscope a larger discrepancy was observed. 1.021 V on the AFG31052 against 960 mV on the scope. A difference of 5.97 %.
I then tried the same measurement with a through terminator on the cable rather than the 50 Ohm termination within the oscilloscope. This showed 1.14 V on the function generator and 0.980 V on the scope. A 14 % difference in the values
I then repeated two of the tests using a 6 metre long BNC cable, that I would usually use for monitoring air gap search coils on generators. The InstaView setup certainly provides a test function for BNC cables. The 1 Ohm termination test showing a 1.68 % difference between the values and the 50 Ohm termination showing a 3.1 % difference.
| {gallery} 6 metre cable tests with InstaView |
|---|
6 metre cable propagation delay test |
Square wave measurement with 6 metre BNC cable and 1 MOhm termination |
Square wave measurement with 6 metre BNC cable and 50 Ohm termination |
As part of the rise and fall time tests required the use of an attenuator, I also looked at how InstaView would respond to this test setup. This test signal was a 10 Vpp square wave and displayed on the oscilloscope showed a measurement of 19.6 Vpp against the 18.53 Vpp displayed by InstaView, a difference of 5.8 %.
It was also noted that the values displayed by InstaView were partially hidden by the trace, making them harder to read, and no resizing of the screen corrected this. This was the only time that I noticed the issue.
With the attenuator in circuit a large discrepancy was seen between the scope and the function generator. An attempt was made to recalibrate the propagation delay with the attenuator fitted to the cable but this produced no delay value and did not correct the difference. This is probably abusing InstaView a little to much though.
InstaView is certainly an interesting aspect of the function generator, although with these initial tests, it seems that one needs to know its limitations before putting it to use.
Square wave rise and fall times
The rise and fall times of the square wave are specified within the verification tests from Tektronix. A 10 MHz, 1 Vpp square wave is used with one reading taken through a 50 Ohm terminator and no attenuation, and a second reading with the 10 dB attenuator in circuit. The test is carried out on both of the output channels. The 53220A53220A counter will only measure rise time, so for consistency, the MDO34 was used for both measurements, the 10-90 % ratio was used for both measurements.
The specification for the AFG31052 is less than 5 ns for both rise and fall times. This is the same as the waveform generator on the MDO34. Both these instruments meet this specification, but none of the other instruments are anywhere near. In fairness, their respective specifications are also no where near that of the AFG31052, this would be another aspect for paying for a higher priced instrument if rise and fall times are more critical for the application.
Only the DG1022 failed to meet is own specification for both the rise and fall times. Since all the other instruments tested okay, and the MDO34 is calibrated, I think it is safe to assume that the DG1022 is slightly out of specification, but then it is over 20 years old and never been adjusted as far as I am aware.
Who Knew?
I seemed to have gotten a little carried away with this blog and have had to split it up into 2 parts due to the following error;
No doubt there are a few on here that have seen this before me, but I will know better for next time.
Remaining tests and conclusions are in part 2.