RoadTest: RoadTest Review the Keysight E36103A
Author: Gough Lui
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?: Various similar power supplies from Tektronix, Keithley, GWInstek, BK Precision and Aim-TTi.
What were the biggest problems encountered?: Design seems to preference rack mounting convenience over front panel ergonomics, and fan can get a little loud and "hunt" at different speeds under load. Hybrid design appears to have higher voltage ripple on the output (~30mV). BenchVue Mobile was a little less functional than expected.
Keysight Technologies E36103A Power Supply RoadTest for element14
It has been another honour and privilege to have been selected to do another substantial RoadTest for element14. My thanks go to Keysight Technologies and element14 for sponsoring, arranging, and administering the RoadTest. As usual, I have spent a good (nearly) two months with the product performing various experiments to produce this impartial, independent, comprehensive and detailed review. It’s a long document, but I hope you all enjoy it - please don’t forget to rate, like, bookmark or leave a comment, and I will do my best to address any potential queries.
As a disclaimer - I am a hobbyist by nature, and it is necessarily the case that the equipment and techniques that I use are not of the high standards you may expect from a qualified testing laboratory following very specific test protocols. As a result, I may report absolute figures which may deviate from the expected figures. This may not conclusively indicate any fault in the equipment, merely a difference in observed result which can arise from different test conditions (e.g. real life "dirty" mains power with variations inside a room with no temperature control), different test equipment accuracy and improvised test protocol. Please keep this in mind as you read through the results.
Lab bench-top power supplies are probably one of the less exciting pieces of equipment as their function seems pretty obvious and uncomplicated, but nonetheless, they are the critical workhorses by which electronics service and testing is performed.
This RoadTest focuses on Keysight Technology's recently announced E36100 basic DC power supply family. This family encompasses a range of power supplies with approximately 40W power rating (6V/5A, 20V/2A, 35V/1A, 60V/0.6A and 100V/0.4A) which are amongst the smallest on the market with a 2U 1/4-rack form factor. These supplies feature USB and LAN (LXI Core) connectivity as standard, with a high-contrast green OLED graphics display, and inbuilt metering that includes low-current measurements. It features a full suite of overvoltage, overcurrent, and overtemperature protection and is suitable for direct front-panel usage on a desk, as well as via remote control and in rack-mount automated scenarios. The supply is supported by BenchVue, Keysight's own "integrated" lab bench software, with basic features for free and other features by paid-add-on-licenses. Remote control is also available through BenchVue Mobile, which is available for Android and iOS. The unit also has IVI drivers for use with third party software, and Keysight I/O Library support for programming your own software. The unit is backed by a 3-year warranty.
The specific model reviewed was the Keysight E36103A (20V/2A). Full specifications and options are shown on the Keysight website for the E36103A, along with software downloads at http://www.keysight.com/en/pd-2585379-pn-E36103A/dc-power-supply-20v-2a-40w.
Basic specifications include a programming accuracy of 0.05% + 7mV voltage and 0.05% + 1mA current, 30mV peak-to-peak/2mV RMS voltage ripple and 1mA RMS current ripple, and a readback accuracy of 0.05% + 5mV voltage and 0.05% + 1mA current.
Before we move on to the unit itself, I think it’s important to first take the approach of a prospective power supply purchaser. If you were in the market for a power supply, you would probably already know what range of voltages and currents you were interested in, what level of programmability in terms of interfaces and resolution you require, and what level of regulation you need. Instead, faced with a product in hand, I approached this in the reverse fashion – namely, what other supplies on the market from leading names have the most in common with the Keysight E36103A – a voltage range covering 0-20V, current range covering 0-2A and programmability with a resolution close to 1mV/0.1mA.
From a brief survey of the market, and with pricing from element14 Australia (as of 26th June 2016), the following comparison matrix of somewhat similar products from leading manufacturers was developed:
Disclaimer: As with all product comparisons, it is impossible to account for every single competing product, specification, advantage or feature of the product in such a simplified comparison. It’s also sometimes impossible to compare products fairly based on the difference in specifications provided. Occasionally, there may also be an error in reading the specifications – so I encourage readers to independently examine the specifications where it concerns them. This comparison is only provided for convenience to discuss the present state of the market.
When compared with other products on the market, it seems that the Keysight E36103A is a little unique. It features a single range output of 40W which is smaller in power than most competing supplies, a few of which feature dual-range features to optimize their value-for-money. It has a comparable programming resolution with most other supplies, whereas the GWInstek PSM-3004 falls behind, and the BK Precision 9171 was worse on current. When it comes to line and load regulation, it is slightly behind its peers in voltage with 1mV worse on-paper specifications as compared to the BK Precision 9171, Tektronix PWS4205 and Keithley 2200-20-5. However, when it comes to current regulation, it surpasses the Tektronix and Keithley offerings, but falls behind the BK Precision, GWInstek and Aim-TTi products. This pattern seems to be repeated for transient response behaviour.
The stand-out difference in specifications relates to ripple voltage, where the Keysight offers 30mV p-p performance which is more in the ballpark of better switch-mode supplies, which is not particularly spectacular when compared with the 3mV or less of their competitors. This is probably a result of the “hybrid” linear design approach used to maintain the small footprint.
The Keysight product does have its strengths. Its emphasis on small footprint results in a 2U, ¼ rack form factor which isn’t duplicated by any of their competitors. It also has USB 2.0 (FS, TMC488) and LAN (LXI Core) connectivity as standard, with an extensive suite of software available – BenchVue (basic free, additional by add-on licenses), I/O Library, Command Expert, IVI & MATLAB (32-bit) driver. Some other products omit LAN functionality or have it as an added cost extra, but other products also feature GPIB options which may be important in “legacy” applications which are not offered by the E36100-family. Other products tend to have more “specialized” single-purpose application software that does not integrate into a workbench. The Keysight product also features a brighter, more contrasty, green graphics OLED display providing better ease of use compared with the older style LCD, LED and VFD front panel displays on competing products. However, its refresh rate may not be as good as VFDs, and the size of the display is necessarily limited by the size of the front panel itself.
The comparison does not take into account features such as programming speed, readback metering range and accuracy for which the Keysight E36103A is said to be one of the leading performers.
Finally, when it comes to the bottom line, the Keysight E36103A was the cheapest of the units selected, listing at AU$1229. Given its unique combination of features and form factor, it’s hard to find any power supplies which are directly an apples-to-apples match for the Keysight offering. As a result, it’s a bit of a toss-up. I’d have to say that each product has its own market appeal, and the Keysight seems to offer an at-least-equivalent value proposition.
I was pleasantly delighted, when a courier knocked on the door unannounced with this particularly “long” cardboard box under his arm, bearing the Keysight Technologies logo on the side.
The items inside were securely packaged, as usual, and survived the international transit unharmed.
The box on top, I would have presumed, would have contained a power cord and possibly some accessories such as fuses, manuals, quick-start guides etc. But instead, when opening it, it was empty. I’m not sure whether this is normal for end-user units – maybe they just included the box to ensure the outer box had some more internal support.
The power supply is safely ensconced by the foam bubble wedges at each end.
Very little unboxing is actually required, as the product isn’t even covered with a plastic bag. Included in the box is the calibration certificate.
The front panel of the unit was smaller than I had expected when looking at images online – it was roughly the size of the palm of my hand. It had a protective film covering the display cover, and all of the banana plug and binding post terminals were front-mounted. Four multi-function rubberized push buttons and a multi-function selector (rotary and push to select) were on the right side, of the same quality as you would find on oscilloscopes. A single hardware power button was on the bottom-left corner.
The top of the supply featured a plastic-coated carry handle, although the end-caps were slightly off from straight. On the whole, it wasn't too heavy to carry which was a surprise - it measured just 3860 grams on my small scale. The unit definitely feels like a quality item judging by the heavy-duty security-panel style cold-rolled steel body which has no flex in it whatsoever.
One side of the unit featured ventilation holes for air intake, through which you could admire the side of the toroidal transformer used inside. The use of forced air ventilation using a cooling fan does raise the prospect of having dust accumulate inside over time, thus needing a clean, and it also raises the possibility of acoustic noise with the unit in operation.
The rear of the unit housed a fused IEC connector, pre-fitted with an appropriate fuse. In my case, the supplied option of O3E indicated pre-configuration for 230V operation. The rear also has the single variable-speed cooling fan, as well as the USB and LAN connectors. A locking slot for use with a Kensington lock is also provided. An internal insulation sheet can also be seen through the case gap. Unlike old-fashioned linear supplies, there is no large heatsink at the back.
The unit stands slightly tall on thick rubber feet, which do a good job of elevating it above the bench (so components can roll around underneath ... as well as the spilt morning coffee), and keeping it from sliding around too much.
In case of relocation, configuration for line voltage is performed through two switches on the underside of the unit. The coding for the switches is shown on the table underneath. A fuse change may also be necessary in case of conversions to and from 230V.
Because of the very few parts involved, the unboxing process was very quick and simple. As Keysight had a teardown video, I didn't feel the need to tear the unit apart. However, curiousity got the better of me, and I did take off the top cover only to find the cover was such a tight fit that it wasn't easy to get it back on again. As a result, I didn't proceed any further, as I feared I might cause some real harm to it if I did.
Setup of the unit for standalone usage entails plugging it in, and pushing the hardware latching power button on the front panel. After a short three-second splash screen, the unit is ready for operation, with no "long" self-test or fan test sequences to speak of.
The unit boots into a default configuration with the outputs off, which is the preferred way. A blinking cursor indicates the setting and decimal place being edited with the multi-selector dial. Pressing the multi-selector dial inwards changes the decimal place, whereas changing the voltage/current setting requires pushing the first grey button to toggle between the two settings. This is acceptable, but not as ergonomic as having two dials. Below the set-point listing is a row reserved for arrows which indicate the status of various protections, errors in SCPI communication, remote communication status and locking of the front panel. The front panel has a lock/unlock button which can be used to lock the front panel against inadvertent changes.
The output state can be toggled on and off with the last button, coloured blue. This enables/disables the output, which pretty much comes on quietly without any "relay clicks" that you might hear on some other power supplies. Unfortunately, the button "feels" the same as the other buttons, so it's not easily identified by touch.
With the output enabled, the top part of the screen is used to display the regulation mode (CC, CV or Unreg), voltage sensing mode (2w - internal, 4w - external) and the metered voltage and current. The meter readings seem to update at about a 3-4Hz rate.
The screen is very contrasty and bright, and somewhat easy to read across a bench provided you have decent eyesight. It seems there was a trade-off made in having the set-points displayed in a smaller font compared to the meter readings, but because the screen only measures ~7cm across, even the larger digits are a bit on the small side even with the improved display font compared to VFD/LCD/LED segment displays. While no flickering was visually perceived, when photographed, the "scanning" nature of the display is apparent, and requires longer shutter times to capture.
Throughout my short period of testing, the OLED display seemed to perform flawlessly, without any appreciable burn-in effects even with the same readings on screen for days at a time. In automated/remote usage, there are commands which can turn the screen off to preserve its lifetime as well.
Configuration is achieved using the Menu button and the multi-selector dial. The menu structure is relatively straightforward and logical, allowing the configuration of almost all parameters via the front panel directly, although some parameters are buried a few levels deep. The whole structure looks as follows:
The screen confirms your selections for certain parameters with clear "Change Saved" or "No Change". Some selections only result in readback of parameters - some with smaller text lists, others with large text.
Useful features include the ability to set the OCP delay and OVP levels explicitly, avoiding unnecessary trips or allowing for more "tight" protection for sensitive loads. The ability to use four-wire remote sensing mode allows for compensation for up to 1V per lead of voltage drop, allowing for precise voltage delivery to a remote load. LAN configurability via the menu allows for allocation of static IPs, configuration of DNS and toggling of services, which is useful if you've "lost" access to the supply on the network.
A total of ten pre-set storage positions is available for storing voltage/current settings. They also store the power supply state as well. Combined with the Power-On feature, you can recall a previous voltage/current setting at start-up, or have the power supply start up with the voltage/current setting and outputs enabled. This is potentially quite handy for semi-unattended or single-purpose situations. Recall allows for the settings to be recalled from storage. One deficiency is that the preset names can not be customized, so it can be a little difficult to remember which slot contains what setting. No confirmation is asked prior to storage, so it is easy to overwrite a previously-used storage position accidentally.
The security menu allows for all device data to be cleared to return to factory settings and secure the device for removal or sale. System Info provides information on the installed firmware, serial number and model. Error menu allows for reading and clearing the SCPI error log on the device.
Connecting your load is as simple as using the standard banana plug connectors on the front panel. These accommodate standard (non-shrouded) 4mm banana plugs, but also can handle bare wires with a hole through the centre for securing bare wire.
The output appears to have a small potential and allows a little current to flow back into the supply possibly due to capacitance and bleed resistor. As a result, when connecting something like a battery, sparks and a slight current draw from the battery may be expected.
The supply is capable of series and parallel connections of four supplies up to the isolation limit of a single supply, as long as the recommendations in the manual are followed. The supply has an included reverse-bias diode so as to protect the supply from damage if backfed by other supplies (if they do not turn on at the same time). This diode will, however, short out anything that's reverse polarity connected - so another reason to take care if connecting batteries, as that could damage the diode internal to the supply.
Finally, we have to discuss the elephant in the room, namely the fan. The small fan in the rear of the unit appears to be thermally controlled, as well as load-controlled. Even with the output switched off, the fan can start-up at a low speed which is barely audible. Once the outputs are switched on, even with no load, the fan seems to increase in speed slightly. Under >1A of load, the fan becomes quite audible like a laptop which is struggling to keep itself cool. Closer to the load limit, the fan appears to "hop" between discrete speeds - an annoying high-pitched whine, and a slightly less-annoying drone. A sample of the recorded noise is attached to the review, so you can hear it for yourself.
But it's not only the fan noise that needs to be mentioned. Because of the hybrid design, it seems that there is another noise which can be audible especially when the fan is at low speed. This sounds like a quiet hissing or buzzing, like some switch-mode supplies often have. This isn't particularly loud, so isn't too annoying on the bench - at least, not compared to the fan under full load.
It seems to me that the product was first optimized for its small form factor so as to fit a 2U 1/4-rack size as a priority, and as a result, it has an odd "long" shape which is a little odd on my bench. This also means a smaller fan which is louder when moving the same amount of air, smaller heatsinks necessitating more airflow to achieve the same cooling, smaller screens resulting in smaller digits, less buttons/dials and a slightly more cramped jack-field. As a result, it seems slightly compromised in its ergonomics, but it's not a bad compromise, as it does provide all the important parameters at a glance and all the necessary configuration at a touch. A more thorough consideration of its performance parameters is achieved through the following experiments and tests.
The main difference between a quality power supply, and a poor quality "low cost" supply, lies in the quality of the output in terms of its ripple and noise, as well as its regulation capability. I tried my best to measure some of these parameters using my PicoScope 2205A (from a former RoadTest), although I will note right at the beginning that some of these results are likely not to be truly representative of the performance, partly because I am trying to measure millivolts on a PC-connected oscilloscope with 8-bit resolution and a 50mV minimum range.
The other problem is that I didn't have a proper load to speak of, so I had to use four 8.2-ohm 10W resistors in a two-parallel, two-series configuration to get as close to 40W loading as I could. The problem with such resistors is that they're inductive, so they can interfere with the power supply's regulation.
Ultimately, what's needed is an electronic load with high bandwidth and a high resolution oscilloscope to do this sort of testing properly. (Hint hint! I shall apply for the other RoadTest ASAP!)
Ripple voltage is probably the easiest parameter to attempt to measure. As a control, I had the oscilloscope with shorted inputs to determine whether there was any intrinsic noise or ripple that would creep into measurements - there weren't.
At a zero-load, the ripple was measured to be 21.89mV average peak-to-peak with no discernible pattern.
At about 50% load (1015mA), the ripple averaged 23.34mV peak-to-peak, with a discernible 50Hz component indicating the main spikes were caused by the shape of the mains.
Increased to around 100% of load (1987mA), the ripple only increased slightly to 26.86mV which still remains below the claimed 30mV peak-to-peak specification.
As a result, the ripple voltage is indeed roughly 30mV peak-to-peak and was verified in testing.
To try and quantify current ripple, a different test resistance of 0.11 ohm was made using 10 x 1.1ohm carbon film resistors to reduce inductance, on the fear that inductive loads may cause unnecessary output ripple. This resistance was run in constant current mode at 1.8A (to avoid over-ranging the oscilloscope) and the derived current ripple was calculated.
The current ripple appeared to be 82.54mA peak to peak on a 1.8A current (~4.6%) which appeared a little high, and the RMS value of the ripple was 6.386mA which is above the claimed 1mA figure. This may have been because of accuracy of the small signal measurement with the Picoscope which may have been susceptible to external interference, and with limited resolution as well. It may have also been a result of the low output voltage in the test. However, the spikes in the current appear in synchronism with the mains frequency, thus it may actually be caused by the hybrid regulation scheme.
Load regulation is quite an interesting parameter, and I proposed to try testing this in the simplest way I know how - use the string of resistors above with a switch over one pair of resistors to "short them out" resulting in a doubling of current in a step change.
While good in theory, tested in practice, I got a whole manner of issues including contact bounce of the switch, and ringing due to the inductance of the wire wound resistors. No firm conclusions could be made, with the exception that in my case, it appeared that the regulation was extremely fast (~0.5us), although the magnitude of the transient appeared somewhat significant as it could not be damped completely by the power supply and doesn't really represent the regulation you would see on a purely resistive load.
A competing manufacturer (who has already been named) claims a feature of their power supply is the ability to protect LEDs and sensitive test devices against turn-on surge currents which could compromise their lifetime or cause them to fail outright. Knowing that I was a complete idiot for trusting a cheap power supply, I had hooked a 5mm LED once to a power supply set to 20mA current limit and ~5V output only to have the LED blow right as the output was turned on. As a result, I know it is a real problem especially with some older units.
To really give the Keysight E36103A a challenge, I set the output to 20V, 20mA, turned the output to off, connected a 5mm LED across the output in series with a 1.1 ohm resistor to monitor the current using the oscilloscope. I toggled the output with great hopes.
As it turns out, the Keysight E36103A managed to reach the current without any transient overshoots, and thus did not cause any damage to the connected LED. The ramp-up in output took about 11.5ms, which is not "immediate" but is much faster than the 50ms "up" program settling time claimed.
With this, came a desire to check how quickly the supply manages to ramp voltage up and down. I chose to use 2V as the "low" state and 18V as the "high" state, thus remaining well within the ranges of my oscilloscope.
Under no-load, changing the voltage up was observed to be completed in about 18ms, much faster than the 50ms spec.
The down transition took 82ms under no load, which is longer, but well within the 150ms spec. Of course, the datasheets will claim larger values to account for all operational conditions and possibly to be slightly conservative.
It was observed, however, that setting the current limiter conservatively at 20mA resulted in the time to ramp up the voltage increasing to 172ms. This may be because of lead capacitance causing some charging current flow, or more likely, it's due to the supply being slightly conservative to avoid over-current in the output.
While I still have the Tektronix PA1000 Power Analyzer from a previous RoadTest, I thought it would be a good idea to examine the power consumption of the supply. Of course, nobody buys a bench-top power supply for its efficiency, but I felt it was still pertinent to check just how efficient this supply was.
The supply was recorded as consuming 8.04W idling with the output off, LAN connected, screen on. This is a significant amount of power, and is probably due to the consumption of the microcontroller, LAN interface, screen, input stage and internal fan.
With the output hooked to my load, running at CC 2A load, the delivered power was 32.336W for an indicated power consumption of 87.29W from the wall. This results in just 37% efficiency for this particular load (close to the 40W maximum). This is not particularly high efficiency, which may go in part to explain why the fan was working so hard to keep the unit happy, as almost 55W of heat was being dissipated within the unit itself.
With other loads, the efficiency may vary, although with no easy way to simulate a wide range of loads, and insufficient time in my day to manually vary the voltage in CC mode to a resistor bank, I left it at that. Nobody buys a bench-top supply on its efficiency anyway.
One of the key features of buying a Keysight power supply is the software support it comes with. The full library of software weighs in at about 1Gb and is free to download. For the purposes of testing, BenchVue 3.5 (17/06/2016), Command Expert 1.6.327, IVI-MATLAB Driver 18.104.22.168 and Keysight I/O Library Suite 17.2.20818.0 were used.
When it comes to software, BenchVue is the star of the show. This is how Keysight tries to "add value" to having a Keysight-based bench by offering a software which aims to maximise functionality with the test equipment and integrate them together to perform testing. The base software is free, which includes the base version of the DMM module which has a 1-hour data-logging limitation, and no threshold alert functionality. Modules for many other Keysight instruments are available, but many of them are not free.
Of course, there are a few other options (e.g. NI LabVIEW, MATLAB/Simulink) when it comes to bench-automation software, the most popular not having such vendor lock-ins and offer improved versatility and flexibility, however, they are not as simple to master and come with significant cost. As they were not readily available to me in the course of the review, I didn't opt to test for compatibility, however as IVI drivers are provided and SCPI commands are supported, it seems almost assured that the unit is compatible with third party software options as well.
The first step when starting up is to connect to the power supply and launch its respective bench application. This process is a little slow, and it takes about half a minute to complete the process.
From there, you can change the settings of the instrument in the Instrument Settings tab.
You can also invoke the BenchVue Test Flow system from the Bench screen, and run a Test Flow with live instruments if you are appropriately licensed. More about Test Flow in the next section.
The Data Logger tab allows you to configure logging parameters, as well as the plotting scale. Data logging can be commenced with the start button.
The side panels can be collapsed to increase the viewport dedicated for the graph. Mousing over the graph shows a cursor which gives you recorded data values. The bench layout can also be divided amongst several other benches where other instruments are used simultaneously.
The Apps tab allows you to manage the installation, licensing and updating of the BenchVue apps. Apps are only applicable for their respective instruments, and many of them are not pre-installed.
The Data Manager tab allows you to view previous data logs, with filtering ability.
These logs launch in their own data log viewer, which preserves the configuration data along with the data.
Notice how viewing a Test Flow data log results in a slightly different viewer that preserves the Test Flow information alongside the data.
The data can be exported in a variety of formats for further analysis. For example, Excel exports look as follows.
The final tab along the top is the Library, where some examples and documentation are provided to help get you started with BenchVue. On the whole, I found BenchVue quite self-explanatory and easy to use, so I didn't have to refer to this at all, although you might find some of the examples helpful.
There are also a variety of settings which can be accessed via the "cog" icon. The first tab was only used to set the language, so I didn't bother taking a screenshot of that, but the second tab is used to set the e-mail settings used for delivering mobile connection requests and alerts.
The third tab is used to set a password and enable BenchVue Mobile access.
The final tab controls the app's privacy and auto-update notification settings.
From the cog dialogue, it is also possible to select the Mobile window to gain access to a QR encoded URL which allows for a phone to connect to the instance of BenchVue on a local network just by scanning the code.
Further details on BenchVue Mobile is provided in a later section.
The next most useful piece of software is the Command Expert. This software supports a multitude of devices with a large range of command sets and can be used to work with other vendor equipment in a limited way.
This is a simple command sequence builder, which doesn't seem to have any flow-control (loop) capabilities, but serves as a useful interface for several reasons.
From Command Expert, it is possible to build sequences which perform certain tasks in a graphical manner with interactive help on required variables. This allows you to understand how to perform a task with an instrument without needing to refer to an old fashioned SCPI command set reference.
With these sequences, you can also profile their performance to determine how long commands take to execute.
In my case, I wanted to know how long it took to measure voltage and current sequentially, through USB and LAN to see whether there were significant performance differences.
In all, no significant performance difference was found, but it was determined that it could take over half a second to retrieve voltage and current readings. It seems that current readings take longer on the supply where it is a low current (e.g. no load) probably because it requires a longer successive approximation to convert the low signal into a digital value.
Command Expert can be used with Excel as a plug-in to run a sequence and return data into a cell or set of cells, although I didn't explore this in too much depth, as it didn't seem as intuitive as I would have liked.
Other than that, Command Expert allows you to generate code that performs the task in a variety of languages, which serves as a good foundation for programming to talk directly to your instruments. Supported outputs include C/C++, C#, CSV/Excel, MATLAB, SCPI and VB.NET.
The I/O Library Suite is the VISA layer and associated drivers that allows for connection to devices and basic diagnostics. At its heart is the user interface known as Keysight Connection Expert. For example, here is the instrument, detected on both LAN and USB interfaces.
This seems to be a required installation for using Keysight's other software packages, which can be a little problematic if you have "other vendor" hardware requiring a competitors' VISA layer to be installed. While Keysight offers to install its own VISA as "secondary" in a side-by-side installation, I found this to be problematic as the "other vendor" software failed to operate with the originally installed VISA which remained installed and instead hung whenever it attempted to detect devices. The device intended for use with the other VISA appeared under Keysight Connection Expert, and despite configuring it to use the other VISA, still remained stubbornly under the control of Keysight's VISA. Ultimately, I had to "transfer" the "other vendor" hardware to a secondary computer to retain functionality - such problems may also be the fault of the other vendor's software as well, but intercompatibility/interoperability issues seem to remain.
The connection expert does allow you to perform some basic SCPI command checks to verify your instrument connectivity.
It also lets you play with some of the other commands which can be used to turn off the screen or make it display something of your own choosing.
With the software installed, the USB connection and LAN are pretty much plug-and-play.
The package seems to claim to have MATLAB drivers for instruments as well, although I believe it was also noted in a Readme that this was for 32-bit installations of MATLAB. As I don't have one at present, I wasn't able to verify this.
You might think I have something against the humble 5mm LED, but I don't. The only reason they were chosen for this test was because they were in plentiful supply in my junk box, and they're so cheap that I could afford to (literally) blow a few. After all, I had no idea what the absolute limits of these "generic" Chinese high-intensity LEDs were, and I could probably learn a thing or two from doing it.
It also gave a very good excuse to fire up a free 30-day evaluation of BenchVue Test Flow to see how it can be used to automate testing.
The Test Flow system works without any coding. It is a graphical "block-based" algorithm designer, with blocks pulled from the left side panel and cascaded to form the test flow. Blocks are "indented" to represent the command grouping in a conventional loop construct, and are colour coded to make identification easy.
During a test-flow execution, each block being executed "lights up" and then fades out, giving a visual cue as to the status of execution. Common errors such as setting output values with the outputs remaining off trigger warning messages. Flows can be debugged by single-stepping through the functions as well.
Test flows log any data and plot it in real time during execution. In our case, as we wanted to plot an I-V curve, we prefer an X-Y chart with the variables set to Measured I vs Measured V, which is easily achieved.
Much more complicated test flows can easily be built - I had a chance to do this, although it is not pictured above. The rate of execution of the flow depends on the response speed of the instrument - in this case, a 5001 loop voltage and current measurement easily took 20-40 minutes depending on the current level (as the returned value takes longer for lower current levels).
It was discovered that during the low current region that the power supply likes to return ~306uA which linearly reduces as the voltage increases, to a value of -38uA for a "virtually zero" current. This seems to imply there is some level of "systematic" error with low-current readings. Data was exported from BenchVue for plotting in Excel.
In all, five LEDs of different colours were tested to determine their I-V curve right up to 5V "destruction" and plotted to determine their forward voltage at the rated 20mA current.
I managed to get a few charred LEDs where the lens itself seems to have melted and burnt silicon smoke was probably ejected along the legs. In other cases, molten silicon seems to have travelled up a wire bond. The not-totally-destroyed LEDs suffered severe semiconductor layer damage resulting in low to non-existent output despite consuming reasonable current. It was insteresting to observe the failures as they happened, as all reached a peak brightness just before the current started dipping, which probably was "runaway" heating of some sort and possibly rapidly decreased emission efficiency. This was followed by "pulsing" or strobing where the LED would regain brightness temporarily and current levels jumped around - possibly due to semiconductor areas that had shorted out being "burnt" open again and not consuming current. Then, almost all output would disappear as current increased further, with occasional "spikes" in current as shorts formed and then were cleared by burning away. Rather interesting, I thought.
However, this is a simple demonstration of how Test Flow can be used to easily create an algorithm using a graphical user interface to automate testing.
One of the touted features is the ability to control devices on a mobile phone using BenchVue Mobile. This is a small (1.6Mb) free app offered by Keysight, requiring Android 2.3.3 or above (or iOS 6.0 or above) and currently at version 2.02.
Once setting a password on BenchVue running on a PC, and enabling mobile access, BenchVue Mobile can connect to the running instance simply by scanning a QR code displayed on screen, or entering the IP address and password. This typically works best across a local network, as WAN access will require port forwarding to operate correctly.
The app allows you to connect to BenchVue on a PC, run a demo or get help. Connecting to the PC lists your instruments.
From here, you can select an instrument to view its associated bench, which allows you to see a data log in progress, stop/start logging and change the graph scaling. The logs are recorded on the PC, and it doesn't seem there is an option to share the log on the phone.
You can also control the instrument's output from the "spanner" icon.
While this is considered "mobile access", its functionality is very limited especially by comparison with their Handheld Meter Logger software where the mobile app actually connects to instruments directly without an intervening PC. With the LAN based capabilities of the power supply, this should be a possibility. Further to this, there's no access to more specialized features such as Test Flow, or reading previous logs.
As a result, it's merely a very simple remote control app, with an old style, which doesn't take much resources. It has its uses, but I suppose I would find it more featureful to use a remote screen-sharing style app to control BenchVue on the PC directly rather than via a limited app like this.
One key drawcard of LXI capable LAN devices is that it avoids the mish-mash of proprietary Ethernet interfaces on test equipment that have a plethora of different interface requirements. LXI specifies several features, of which a web interface is one of them.
Just by accessing the IP-address, you are presented with all the information about the device.
With a login (default: keysight), you have full access to the instrument through the web browser - whether it be a desktop or a mobile web browser.
This includes front panel control and viewing of the display ...
... the ability to download screenshots of the front panel in .bmp format ...
... the ability to configure the LAN connection ...
... the ability to reset the network configuration ...
... and the ability to change the password on the instrument, which is probably recommended.
Rather concerning, or at least expected in limited-compute embedded devices, is that all of these functions are available over unencrypted plain HTTP.
A quick check with Wireshark showed that it uses "cookie" authentication, with cgi scripts, basic encoded POST form data and JSON returns.
Without the cookie, the instrument rejects any commands, as expected. But because the cookie is sent "in the clear", it can be captured from a network and "replayed" for as long as the previous user has not clicked the log-out button.
It was also discovered that the /get/preview.bmp front panel image was always accessible, logged in or not.
As a result, it's really important to realize that the LXI LAN interface really has limited protection against network tampering. There are direct access ports, as well as HTTP services which can be accessed without any authentication or encryption, and basic cookie security for the web interface which is vulnerable to "cookie jacking" and replay. I wouldn't advise making the instrument reachable over the internet directly. Where the power supply is hooked up to, and controlling critical safety-dependent functions, having this type of "hole" is not acceptable - so my advice would be either to use USB (as it's a one-to-one connection), or use the LAN interface on a separate network to the rest of your network traffic, or place it on a separate VLAN to prevent interference by regular network users.
As I ultimately knew I would lose access to BenchVue "Pro" features once the trial was up, and I realized that BenchVue and the VISA layer were ultimately hefty applications which might be unwelcome in some circumstances, I realized that it was actually entirely unnecessary to have any of this cruft if we leverage LXI Core's functionality of SCPI Direct. This is Port 5025, which offers a "socket" based interface to the device which you can send SCPI commands directly to.
On an interactive keyboard session, using socat is enough to command the instrument, and quite effectively as well.
Using a little bit of C-knowledge and borrowing from other C-socket and time code online, I made a small C program I called scpidir which allows you to connect to an instrument and log voltage and current pairs by specifying the IP address, port and logfile name.
The source is attached to the review, and can be simply modified for other purposes - including many of the investigations performed after this.
This begins a long journey to actually trying to determine how accurate the output voltage of the supply is, under a no load condition (to avoid any loss of voltage in test leads). I used my Keithley Model 2110 5.5 digit DMM as the reference measurement device, set to auto-ranging 10PLC (as 100PLC resulted in greater errors) and recorded the output.
The power supply was commanded with a variation of the above program that set it to each 100mV step from 0.1V through to 20.6V. The first three and last three measurements around step-changes from the unsynchronized multimeter were discarded, and the remaining readings (~260) were averaged to find the mean error, and the range of the values were also determined and plotted on a graph.
Because of the time consuming nature of letting the unit settle at each step and collecting a sizeable number of readings, then processing it, the graph below may look simple, but actually took a long time to make.
On the whole, it shows that the mean voltage error on all steps was within a 1.6mV band, not accounting for errors in the reading multimeter which was maximum 1.4mV. This implies that the mean program error was likely to be very close to zero and extremely likely to be below 2mV which was much below the 0.05% + 8mV stated, likely because of the "easy" no-load condition. The range of voltage variation in each step did increase as the voltage increased, but was itself well within 1mV. From my perspective, this is a great result that challenges the accuracy of my multimeter!
The same exercise was repeated for a 4.1 ohm load resistance load, measuring current accuracy instead at each 10mA step from 0.01A to 2.06A.
Without considering the accuracy of the multimeter, the values were within a 0.6mA range of the programmed value (when considering the range of returned values) which was astounding. When the accuracy of the meter is considered, it was only supposedly capable of resolving within 1.7mA at 1A and 4.9mA at 2A. At lower current ranges, it was capable of resolving 0.06mA (100mA), thus it seems at least the results from the lower ranges can be trusted. It seems likely that the programming accuracy might well be better than the metering ability of my multimeter. Compared with the datasheet specification of 0.05% + 1mA, it was well within just the 1mA part, making it an excellent result.
In all, if we consider these results in isolation, I think we can conclude that the output programming accuracy is extremely low error to the point that it is challenging the resolution of a 5.5 digit multimeter. In light of this, the next segment which looks at readback accuracy will assume that the programmed value is the actual value output by the supply, so we can characterize the metering error.
One of the biggest features touted was the read-back capabilities, especially small-current. I again modified the program above to step every millivolt output while taking the average of 25 readings from the device.
The combined read-out showed that in small voltages, the error reached a little above 3mV, well within the 0.05% + 5mV readback accuracy claimed. The majority of the case, it was within 2mV.
Because the above test took over 24 hours to actually complete, I was a little less thorough with the current, stepping through each 1 mA step instead of 0.1mA step.
This resulted in a maximum error of under 350uA, which is again, below the claimed 0.05% + 1mA. The current error, however, seems to be constantly in the positive territory which implies there may be a slight offset error to the calibration, but well within requirements.
In all, it seems the unit's metering seems to perform well and above the spec which is nice to see.
One potential use of a power supply is to charge a battery. Inevitably, every power supply that I’ve used, sophisticated or otherwise, has been in some way abused to do this. Using a programmable computer-controlled power supply yields interesting benefits of being able to control the charging algorithm and record the voltages and currents over time to get a good picture of the state of the battery and the charge delivered.
It also just so happened that an old neglected Century PS1270S 12v 7Ah AGM Sealed Lead-Acid battery caught my eye from across the room, so I thought maybe it’s time to revive it. As a result, I decided to employ the IUoU (3-stage) charging algorithm, using 2.0A for the CC part, 14.4v for the Uo CV part, and finally 13.8v for the U CV part. Given that the battery had been forgotten for a few years, I didn’t expect much.
The first thing I noticed was that the power supply had some potential on its output even when switched off. It was in the order of 28mV or thereabouts. It also sparked when I connected the leads, indicating there was a current draw of some sort, probably due to capacitor charging. As a result, I’d recommend the use of an external diode and possibly using the 4W (external sense) mode to compensate for the diode drop.
Another warning is that the output of the power supply has a “series-ing” diode installed internally that conducts when reverse polarity is applied to protect the supply if it is switched off and other supplies in series are still supplying voltage. This diode will cause the supply to appear as a near-short to anything connected in reverse polarity, and failure to take care of this may lead to the destruction of the diode or high current flow causing nasty heating.
Being more familiar with my own “simple” socket-based logging program, I decided to log the charging condition over time and plot it in a graph.
Initially, the battery seemed to have its voltage “float” to the CV limit almost immediately on application of power. Maybe this indicates a bad case of sulphation, but soon after, it may have partly broken down and the generated heat internally may have helped the battery accept some current. Very quickly, the bulk phase (U) was completed, and we reached the absorption phase (Uo) where the current fell off over a period of a few hours, but then the rate of decline slowed down. I would have liked to see it reach C/100 (as a healthy battery will), but it didn’t achieve this even after 13 hours of absorption, so I switched it over to the float phase (U).
At this point, another interesting characteristic of the supply surfaced. When the voltage was lowered on the supply below the battery voltage, the supply acted as a small load consuming about 24-28mA according to the front panel. I would have expected that there would be no load at all and the supply would let the voltage drift down on its own accord. This consumption may be reflective of internal capacitor bleed resistors or similar, used to discharge the capacitors especially on downward voltage transitions to ensure good “no-load” voltage program speeds.
The float current did slowly decrease, so the battery wasn’t too bad. A really bad battery would have a high float current that persists with no sign of tapering off. By charging the battery using a supply with a good metering ability, we could both tailor the charging algorithm and observe and record the charge progress.
The next little experiment exploits the 2A rating of the power supply to act as a USB power supply (using a USB A-F plug) for examining the current draw behaviour of an Anker PowerCore 10050+ power bank. The current draw characteristic was previously tested with the Keysight U1461A with a USB shunt resistance and a standard wall charger. The logging code presented earlier was used to collect the data.
Whereas the previous measurement only showed the current, the integrated metering of the supply allowed both voltage and current to be observed without any fear of shunt resistances impacting on the delivered power.
It allowed us to observe that the power bank was drawing fractionally over the nominal 2A input current at times in very short bursts because of the switching converter within the power bank. This gave the power supply a good workout, as it was alternating rapidly between CV and CC regimes to ensure that 2A was not breached. It was so strenuous that for a very short period, "Unreg" displayed on the front panel, as the regulation was caught "in-between" the regulation regimes.
It clearly shows that the power bank was drawing peaks of 2A hence the voltage reduced, but also seems to allude to a "source resistance" sensing strategy which is used by some heavier load USB equipment to determine the current capacity of the supply it is connected to - the voltage drops in response to overload, and the device "backs off" in its consumption.
This got me to the point of thinking - what applications besides the I-V test performed earlier might benefit from tight voltage regulation, and Lithium-Ion battery charging appears to be one of them. For safety reasons, and for the best lifetime of the cells, the CV portion of the charge strategy must be maintained at a set voltage (commonly 4.2V, but 4.3/4.35V in some higher capacity cells). A few tens of millivolts either way will yield slightly reduced capacity or significantly reduced lifetime and potential for safety hazard. You won't want to be charging a Li-Ion cell on an "unknown" supply.
As I had a lot of "scrap" salvaged 18650 cells of various vintages in batches from old laptop batteries, I wanted to determine their capacity. Of course, this is traditionally done by a load test (but I didn't have a suitable load ... so I'll try to get one), but if we use the simple fact that the colombic efficiency of Li-Ion is 100%, we can determine the capacity from charging a fully discharged cell.
By using a standard appliance with a fixed voltage cut-off and a fixed load, a total of twenty cells were first discharged. They were then charged with the appropriate strategy (termination voltage, constant current limit, and termination current) as specified by the manufacturer, or a "generic" conservative 4.2V/750mA/100mA strategy where information was not available. The logging code presented earlier was used to collect the data.
The current and voltage behaviour of the cells can be seen in the graph above. The Panasonic cells were the latest harvest and were in very good condition, having a good CC portion and a reasonably controlled CV portion for a relatively "fast" charge under 2 hours 30 minutes. The Samsung cells seem to have a greater internal resistance, so they terminated their CC portion earlier, despite receiving a lower current, and spent a much longer period in tapering off, charging in about 3 to 3.5 hours. The remaining cells were charged at a conservative current level, and shows that if using a slightly lower CC level, the charging can complete in similar times as a higher CC level, as the cell spends more time in the taper regime.
The portion where the capacity was determined was done using its internal metering, by integrating the current over time down to the cut-off current level. This resulted in a capacity table as follows:
On the whole, cells from the same batches (in different colours) showed very similar results - in many cases within fractions of a milli-amp-hour. Some cells showed drastic differences, namely the "unknown" no brand cells. The branded cells typically were more consistent, although the first Panasonic seems to be a bit of a "stand-out" cell.
Computing the integrated mAh error based on the datasheet, it was determined that the results were known to within 4mAh, which means that the metering is quite suitable for power monitoring purposes.
I think you'll agree that this has been a long RoadTest, but it's been a great experience working with the power supply, and it has taught me as much as I have learned from it.
The Keysight E36103A is a 0-20V/0-2A programmable power supply with a hybrid linear design that features a sturdy physical build, small size, rack mountability, sharp OLED display, good software support and USB + LAN connectivity as standard. The unit is backed up by a three year warranty. The power supply compares well on cost, size, speed and metering capabilities. It comes very close to its peers with regards to regulation performance. Its main drawback appears to be a significantly higher output ripple voltage, which is closer to switch-mode supplies at 30mV and the lack of "dual range" versatility.
Its small size and long body makes it a little out-of-place on a bench, and it seems they have prioritized small rack-mounting form factor above ergonomics, as the display might prove a little small for some users, and the jacks a little close to each other. The single control wheel and lack of tactile feedback on the buttons also serve to frustrate some users. Above all, the fan noise was quite distracting at high loads, especially with the "hunting" between speeds which occurred at the high end of the load range.
The included connectivity and software was a great bonus, functioning flawlessly throughout testing and offering basic logging abilities out of the box. Additional tools to help users integrate the instrument into their own programs were also quite helpful. Evaluations of the test flow feature showed that it is easily possible to constructed automated test algorithms without any coding, using a visual GUI interface that was attractive and intuitive. During execution, it also very clearly visually represented the program flow and steps, making it quite entertaining and informative. Unfortunately, full functionality requires licensing BenchVue Power Supply Pro for notifications and unlimited logging, and BenchVue Test Flow for automated test flow features.
BenchVue Mobile, however, appeared to be a slightly neglected simple remote control application which was very limited in functionality. You may be better served with a general purpose remote screen-sharing app instead.
The LAN connectivity was simple to set-up via the front panel and offered very similar (indistinguishable) performance from USB. LXI core connectivity allowed for direct browser access to the instrument on both desktop and mobile, and control of the front panel. It was discovered that the security of such systems is rather limited owing to the SCPI direct port (5025) being open to all without authentication, the use of unencrypted HTTP connections to the instrument and the use of basic cookie "authentication". I would not advise using the LAN connectivity on untrusted networks, such as the internet.
Some difficulty was encountered in trying to characterize the regulation capabilities of the supply, however, it was found that in most cases, the voltage program rates surpassed that of the claims in the datasheet, the voltage ripple was consistent with the datasheet, and the current ripple was a little higher than that of the datasheet. Testing of overshoot in constant current operation showed no issues.
The program and readback accuracy was very good, rivalling the ability of the 5.5 digit DMM used to test the unit. The results were well within the specifications, <0.2mV and <0.6mA of error observed in all cases.
Given its position in the market in regards to price, it seems to be a well-specified "basic" affordable programmable power supply which compares closely with its competitors with its own specific advantages and appeal - namely its OLED display, ease of use, small size, metering accuracy and software support.
Thanks for reading – don’t forget to like, bookmark, rate or leave a comment. I’ll try and answer any queries as soon as I can, and as best as I can. Thanks again to Keysight Technologies and element14 for arranging this RoadTest and entrusting me with the product. Rest assured, it will be put to good use as another one of my faithful bench-companions for future experiments and investigations.
For the latest on what I’m up to, feel free to visit my personal blog at http://goughlui.com
Thanks for that piece of information - I suppose that firmware will affect the behaviour of the device, but I had (incorrectly) assumed that the firmware I received was the "initial" base that everyone started off with. Evidently, you've managed to get a unit with firmware even older, so missing features and bugs might indeed be expected. I'm glad you were able to solve it, and this information will probably be quite helpful to others who might have obtained a unit with an early firmware version.
We were able to get it to work! We couldn't even connect on LINUX with netcat, and the port scan still showed nothing on 5025. The last thing we checked was firmware and our instrument was running 0.3.3-0.33.
I looked thru your review and found a picture showing your instrument was running 1.0.0-1.00. I was able to find firmware 1.0.5-1.01. After downloading the firmware (along with USB driver utility), there were missing VISA drivers. Found out that my existing TekVISA drivers are not compatible (literally there was a conflict so we had to remove them), so when I installed the Keysight IO Libraries (includes VISA), I was able to connect with USB and update.
After that, I am getting a response!
Thanks for your help! I probably would have send this PS back to vendor and looked at another option. We can now move forward with using multiple of these on a network bus to control an LED lighting test.
That sounds quite strange, as if you can get a socket connection to the Web UI, I'm assuming your network connectivity is just fine. You might try setting up a static IP on the power supply and on your laptop/computer and use a direct crossover cable connection (or straight through if one end is Gigabit capable as they will auto-crossover).
That is, unless maybe some other device or service has already connected to the port somehow. Maybe you have a VISA implementation installed that has picked up the device and somehow constantly maintains a connection to it? I had no issues with Keysight VISA being installed, but mine was installed secondary alongside a NI VISA installation as primary.
Maybe using Wireshark to capture the packets could be helpful just to see if the TCP connection is established, as it's hard to tell with the netcat/PuTTY, as they will "sit there" for almost 60 seconds before timing out in case of a failed socket open. The device doesn't say anything by default on a socket open either, but I confirm that there are no special configuration settings I'm aware of in the power supply menus that would change its SCPI-over-LAN behaviour.
Thanks for the quick reply, and we'll try what you suggested. Seems strange we can't even establish TCP socket to the device, so we're going to try a 'clean' install and direct connection as I suspect there may be some firewall settings or physical port filtering on our network switches. If that doesn't work we'll fire up a Linux box. Thanks for your help!
I'm not sure what is going wrong with your configuration, as I don't recall having to do anything specific for it to just work. However, that being said, as part of the review, I did tweak quite a few settings which I didn't fully keep a track of. Just when it came to testing SCPI over LAN, it was working perfectly for me.
If the device works over the WebUI, and the address is known correctly, then maybe you need to try using another tool to access it via TCP or check whether the line endings are sent correctly. In the case it sends <CR><LF> (e.g. Windows based) versus just <LF> (e.g. Unix based), behaviour can become unexpected depending on the device including no response at all. As a result, you should make sure your commands are terminated in just <LF> - for example, in Linux based scenarios, your line endings should be \n rather than \r\n.
You will know if you have incorrect line ending issues if you don't see a response from the device, and the ERR announciator on the front panel is flashing. It means that some command has been received but could not be processed due to error (e.g. 102, 103, 113). You can check the error in the Error menu.
If you're in Windows, using Cygwin with netcat, you can issue:
Alternatively, if you have access to any Linux based machine, netcat pretty much comes installed as standard, and it should work just fine. Do not use Hyperterminal - since it always seems to send <CR> with only an option to add <LF>. It will not work. I also cannot vouch for various telnet clients, as they probably will do the same thing.
A more compact client could be to use in Windows would be PuTTY in Raw mode (not SSH, Telnet, etc)
The most basic SCPI command is what I've started with in the examples - namely, the short identify code command *IDN?<LF>
If you see the reply, then your comms is working just fine - it's just your choice of client may not be sending line-endings correctly.
Great review. I wanted to ask a question about the SCPI over LAN. We are trying to connect using port 5025 and cannot seem to get a response from the device. Were there any other settings you had to change to enable this communication? We are able to use WebUI and ping the device from the local network. Any advice on how you were able to confirm or establish TCP connection would help. Thanks!
Very nice review - thanks.
A very thorough road test, thanks for sharing
Nice job, seems you covered almost all the possible applications of this device.