The final detailed blog corresponds to the most difficult part of the review and potentially the most interesting as well – how well does the HMP4040.04 perform? I threw my barrage of test equipment at it to try and find out.
Things to Note
If you have not read the unboxing chapter, it is important to note that this HMP4040.04 arrived in a damaged condition. Rather than delay releasing a RoadTest review significantly, I have decided to release this review of the damaged unit with an understanding that it would be updated once an undamaged unit has been supplied. As a result, testing has been performed with some limitations and some of the observed issues may not be a result of the unit’s design. Readers are asked to keep this in mind as they read the review.
In Ch3, I noted how the unit lost the ability to generate output on Channel 4. In Ch4, I noted that there were further anomalies occurred with regards to PC-connected remote control which could be overcome with a number of hard power cycles. Testing of the HMP4040.04 was undertaken with the current firmware at time of review, version 2.62 on a computer running Windows 10, NI-VISA 18.0, WinPython and pyvisa.
While the unit did have its anomalies, instrument performance testing was actually performed prior to a number of the other tests to identify whether the shipping damage had affected the unit significantly or not. As a result, the results presented here for voltage/current programming accuracy are extremely comprehensive as they were undertaken prior to all other testing.
Insulation Resistance Testing
Prior to using the transit damaged HMP4040.04, the first thing I did was to ensure the insulation was still integral by performing an insulation resistance test using a Keysight at 500V for 10 minutes between various primary and secondary connection points, as well as between the primary and earth.
The lowest reading was about 78Gohms, which is more than adequate to ensure safety, thus the unit’s insulation was judged not to be compromised. Upon plugging in and powering up, it behaved as expected (at least initially).
Voltage/Current Programming and Read-Back Accuracy
Testing of the voltage and current programming accuracy was performed with my Keithley Model 2110 5.5-digit digital multimeter as the reference measurement device. While this multimeter has been technically “out of calibration” for many years, based on cross-referencing with in-calibration devices, it does not appear to have drifted significantly. An pyvisa script was used to automate the test process, which involved setting a fixed voltage/current value on the power supply (at every 1mV or 0.1/1mA step) and averaging 16 readings from the Keithley DMM taken at 10PLC integration time for the best accuracy. The programming error was determined by taking the difference of the DMM value and programmed value, with the read-back error determined by taking the difference readout value and DMM value. Where range changes were necessary, they were accomplished automatically where possible and manually (e.g. 3A/10A) where necessary. The results were plotted, with the blue line representing the difference of raw values, the green line representing the upper boundary and gold line representing the lower boundary of the actual value based on the DMM’s error specification, and the red and black lines indicating the boundary of acceptable values based on the HMP4040.04’s specifications.
Looking at the voltage programming accuracy (left hand column of graphs), it is clear that each channel has its own characteristic, but all channels performed exceptionally well, with the raw value deltas being very close to zero and the boundary of certainty encompassing the zero-delta line. This means that the HMP4040.04’s specifications are very much conservative, and the accuracy of programming is better than advertised.
The voltage read-back accuracy is slightly poorer, with some level of upward bias indicating the HMP4040.04 seems to be over-reading slightly. However, it still remains well within the specification limits, and the margin of error of the DMM does still encompass the zero-delta line implying that the DMM used isn’t quite accurate enough to be sure that the HMP4040.04 is actually in error by any significant amount. Again, another excellent result.
[UPDATE - Due to errors discovered after publication, the error margins in the following current programming and readback accuracy graphs are incorrect. The correct data is posted here]
The current programming accuracy is, perhaps, even more impressive. From the graphs, it seems that the Keithley 2110 in the 3A range is a little bit off at the upper end of the range – this can occur due to the current shunt warming up during measurements, but overall, the current again lies well within the specification boundaries and the margin of error straddles the zero-delta line for the most part, indicating the HMP4040.04 is basically providing the level of current as commanded (to the resolution available on a 5.5-digit DMM). The readback performance is also quite close to that of the programming performance.
Keeping in mind that this testing was performed at home, using “dirty” mains power which has noise, in a room with limited temperature control, over the course of about a week to collect the data, this result is absolutely unexpectedly good and implies that R&S have been conservative about their ratings – inspiring confidence that the promised performance can be maintained in sub-optimal operating conditions.
Channel-to-Channel Timing & Power On/Off Behaviour
While I did mention in Ch3 that the HMP4040.04 seems to lack any channel sequencing/delay features, I wanted to understand how the power rails come up and shut down and their timings.
The first was to test the channel-to-channel skew. At this time, only three of the channels were functional. The oscilloscope was set to have infinite persistence, triggered on the first channel, while the power supply was cycled on and off 500-times under remote control (right). Cursors were placed to determine the maximum time deviation. The result was verified by having all probes connected to the same channel to establish that the RTM3004 oscilloscope’s input channels are practically synchronised (left).
In the test sample of operations, the channel turn-on times varied around a range of 2.73ms. This is not entirely unexpected, as the HMP4040.04 uses relay switching, and the mechanical nature of relays does introduce some variance to the timings.
Power-on rise time of the power rails as configured for 1V, 16V and 32V operation was around 2ms which is relatively quick. With the rails unloaded, the rise time is consistent across the voltage settings with no overshoot.
On power-down, the fall times were about 16ms which is also fairly quick, indicating that there are no large capacitances on the rails paired with an appropriately low-value discharge resistor. The curve indicates it is merely a capacitor discharging.
It was also of interest to investigate what the behaviour of the HMP4040.04 is when power is abruptly removed during operation. Some other power supplies have specific warnings not to do this as regulation can be lost on such a power-down. Looking at the scope traces above, an abrupt power-down using the red hardware power button results in a less-well-defined rail power down timing, along with significant over-voltage on the first rail configured as 1V which shoots up to 10V. As a result of this behaviour, in case of emergency, it is better to use the master output switch rather than the hardware power button to ensure no risk of damage to the DUT.
Constant Current Overshoot
With all power supplies that can operate in constant current mode, there is a short window of time especially during power-on where significant current transients can exist before the supply regulates in constant current mode. This could be potentially damaging and was worth testing.
Unfortunately, testing for constant current overshoot reliably is not an easy thing to do as the behaviour of the supply can depend on the characteristics of your load. In this case, the load is a wire-wound 10W 4.7ohm resistor which is somewhat inductive, which could exacerbate the overshoot somewhat. A fixed current limit of 1.4A is set on the supply, and the voltage is altered. Switching the output on and off, the voltage across the resistor is recorded.
The bigger the difference between the operating voltage (~6.58V) and the set voltage, the larger the overshoot. However, the overshoot only persists for very short periods and quickly settles into the expected voltage within about 4ms. With smaller voltage differences, the overshoot is smaller but the supply takes a little longer to settle into the constant current mode, up to about 7ms. In the case of 32V setting, the peak transient reached about 11.61V.
While it would be great to have a power supply that has no voltage overshoot, it is virtually impossible to achieve and could result in a trade-off in the turn-on or voltage program-up characteristics. For the short length of time that the overshoot persists for, it does not seem to be a major problem in practice provided a sensible voltage choice is made initially.
I proved this to myself when I connected a single 5mm “standard intensity” Red LED to the channel output, dialled the voltage right up to 32V and the current to 50mA and cycled the output multiple times. While the transient does result in a bright turn-on, the LED was not permanently damaged with multiple power up/down cycles. I had previously done this with some lesser power supplies and blown an LED straight away on the first try – so this performance is, again, rather exemplary.
Transient Load Regulation
Testing of the transient performance of the HMP4040.04 was undertaken with the B&K Precision Model 8600 DC Electronic Load. Load steps of 1A to 2A (and back again), as well as 1A to 9A (and back again) were performed and the performance judged by 10:1 probes connected an R&S RTM3004 oscilloscope. Four-wire sensing mode was used for the power supply to the load to minimise any voltage drops due to wires appearing in the output. The power supply was set to 15V to ensure that it could provide the full amount of current without overloading the electronic load.
While the CV load regulation is given in the datasheet as <0.01%+2mV which would imply acceptable variations to be 35mV for a voltage of 15V, the issue is that the slew rate is not specified which makes testing rather difficult. Also, oscillations were discovered in the DC electronic loads’ profile, which makes the actual voltage change somewhat for the lower slew rates. As a result, the experiment was repeated for rise times of 10us, 75us and 1.17ms. The shortest slew rates are able to cause a significant deviation in the voltage of the rail – all power supplies I have tested seem to have this behaviour, but the recovery time is extremely quick.
In the other direction, an overshoot is seen but the behaviour is very much otherwise similar.
Performing an even higher change in current magnitude from 1 to 9A and back again over a gentle 10ms results in a slight reduction in voltage (possibly due to probing on a “voltage gradient” developed at the terminals) but shows fairly good regulation behaviour.
It seems from my informal testing that the transient load regulation is rather agile and accurate, although the DC electronic load was even more agile, thus it was possible to get some rather significant rail excursions. This performance is excellent regardless.
Ripple and Noise
Another key specification is ripple and noise which describes the cleanliness of the output from the power supply. Unfortunately, while I was proposing to test this, I have since learned that ripple and noise measurement is an art (especially when it comes to low levels). While you could hook up an oscilloscope to the output of the power supply, the intrinsic noise level of the probe and scope input combination could cause problems especially when using 10:1 probes. While we can claw some of this back by changing over to a 1:1 probe (which is what I eventually did) using AC coupling on a very sensitive range, the problem with a 1:1 probe is that it won’t be able to resolve higher frequency noise components. Ideally this should be measured with a dedicated power rail probe such as the R&S ZPR-20 Power Rail Probe, but sadly, I don’t have one of these.
With the RTM3004 connected to a 1:1 probe, in AC-coupling mode with 20Mhz bandwidth limiting and the HMP4040.04 providing 16V into no load, the measured output shows an RMS voltage of 12.863mV with a significant 175khz ripple component.
An FFT of this shows that the HMP4040.04 also seems to generate some noise in the higher frequencies around 5Mhz and 7Mhz which is not particularly friendly for radio users.
Turning off the output on the HMP4040.04 does not quench the noise.
But turning off the power switch to the HMP4040.04 kills this off immediately. Unfortunately, there is still some level of background noise – the source of which is not known.
Reverting back to the time domain with the power supply switched off using the hardware switch, the scope still sees occasional spikes reaching into a few mV.
Finally, with the input shorted using the REF switch on the probes, we are still seeing about a 1mV peak-to-peak signal occur. I wonder if this is something generated within the scope itself or received from an emission from the oscilloscope. But knowing this, it seems rather unlikely we could verify a “rated” 1.5mV RMS ripple performance using this set-up and the fact is that the amplitude of the noise shown is likely to have significant contributions from the probe/cable/scope combination itself.
Standby/Idle Power & Line Regulation
As the HMP4040.04 features a hardware power switch, turning the unit off using the red button drops the power consumption to (practically) zero. On a standby power test using the Tektronix and an inverter synthesised pure-sine-wave source at 230V/50Hz, I measured a standby power of 0.5mW with an uncertainty of 1.1mW – essentially zero.
With the supply powered up, but all outputs switched off, the supply consumed a quiescent power of 30.609W (with 0.3W uncertainty).
This increases slightly with the rails switched on, but no load connected. In this case, it was measured to be 37.561W (with 0.332W uncertainty).
This means that the supply does consume a fair amount of power if switched on, even if the rails are turned off. This is not particularly surprising, as the linear toroidal transformer and regulation circuitry itself are likely to consume some level of power. But once the hardware power switch is used, the standby current is zero – so make it a habit to turn off the supply when it’s not in use.
I also tested what happens when the mains voltage is slowly reduced. With no load on the output, the supply seems fairly happy even down to 190V at the 230V setting. Once below 190V, the power consumption appears to increase somewhat but the maximum output voltage on the DC rails begins to reduce.
A message indicating “UNSTABLE!” appears on the supply under this condition, with further reduction in voltage below about 170V resulting in a power failure message and the unit locking up, requiring a restart. Not far below this, the unit starts to cycle on-and-off, flashing its LCD. Testing of the unit with input voltages up to 272V did not result in any anomalous behaviour. This seems to indicate that the HMP4040.04 has been designed to monitor the input power conditions to alert users to issues and is able to operate throughout the rated voltage range with some margin to spare. Unfortunately, due to the limited current handling ability of my isolated synthesised mains supply and variable-transformer, it was not possible to verify performance under load.
I’ve put the HMP4040.04 through my barrage of tests and its performance can only be described as unimpeachable. Despite the shipping damage, the unit’s insulation was uncompromised and the unit was (at least initially, fully) electrically functional.
While each rail is slightly different, voltage programming accuracy was extremely tight with most of the results showing agreement with a 5.5-digit DMM within the margin of error and being well within the claimed accuracy levels. The voltage read-back accuracy was slightly poorer, but still well within the claimed limits. Current programming and read-back accuracy was even more impressive, again being mostly within the margin of error of a 5.5-digit DMM. All of this, when considering the power supply was operating in a non-temperature-controlled room off raw mains power which has noise and voltage variations. This suggests that Rohde & Schwarz have been extremely conservative with their specifications to ensure they will be met even under adverse operating conditions and that the unit was well calibrated from the factory, not drifting much despite the abusive shipping conditions.
Channel-to-channel power-up timing varied by about 2.73ms owing to the relay-controlled output and potential mechanical variances at play, which is still quite acceptable. Rail ramp-up and ramp-down times were a swift ~2ms and ~16ms respectively.
Overshoot into constant current mode was observed, with the overshoot magnitude being greater for a higher voltage, but the recovery is also more rapid at higher voltages. In the tested cases, the constant current regulation was achieved in 4-7ms after the channel comes up – a very rapid response. Testing this, I connected a 5mm red LED to the outputs, set to 32V at 50mA and cycled the output multiple times. The LED survived seemingly undamaged – where doing the same thing to some other supplies had resulted in the LED being destroyed at the first cycle. This demonstrates just how responsive the regulation loop is.
Transient regulation was tested informally using a DC electronic load. While the load was able to provoke some voltage excursions, this is not unexpected at high slew rates. On the whole, the response was very quick and any excursions were very limited in duration. If anything, it demonstrates that the load is faster but the power supply is still very quick to respond.
Attempting to quantify the noise and ripple was more difficult owing to difficulties with ensuring a balance of bandwidth and noise levels. In the end, I was not able to conclude much about the actual ripple voltage amplitude, however, it seems that there is significant ripple at 175khz and also noise at 5 and 7Mhz which might not be appreciated by LF/HF radio loads. This noise appears to be present whenever the power supply is switched on, even with the outputs disabled.
Finally, it was determined that the power supply was able to operate throughout the voltage range required with some additional margin. Internal monitoring is able to inform the user when the power supply outputs are unstable and when the voltage has dipped too far.
The only potentially major issue was the behaviour of the rails when power is removed from the power supply unexpectedly. In the test case, an unloaded 1V rail rose to 10V as the hardware power switch was actuated – so it’s best to turn off the rails first using the output button before turning off the power to the supply to ensure the DUT is not damaged. Another observation was a ~30W power consumption of an idle power supply with outputs switched on. This is a significant draw, but if the supply is not in use, the hardware power button can be used to bring the draw down to zero.
You can download the scripts used in this section in a .zip archive attached to this post, which includes the Python pyvisa scripts used for testing the voltage and current readback accuracy.
This blog is a part of the Rohde & Schwarz HMP4040.04 Programmable Power Supply RoadTest Review.
hmp4040-ch5-scripts.zip 3.1 KB