Korad KEL2010 reviewDidn't want to create a new thread so added my review here.
IntroductionFor over a year I have been looking at upgrading my DC loads. I used to have a couple of loads from Array but these never managed to impress me so I got rid of these. This left me without a compact, bench top DC load. I do have an Agilent N3301 DC load mainframe (with a 120V 60A 600W and 60V / 120V 600W module) but this doesn't have the ability to adjust the current accurately at low levels (tens of mA). It is also quite big. So time to hunt for a new load and that wasn't very easy to find. A feature high on my list is to have binding posts that can also take 4mm banana plugs.
Many DC loads seem to have limited features and a small display and/or are plagued with firmware issues (even the ones from Keysight don't give me a warm & fuzzy feeling based on the reviews). In the end I found two models that seemed to be at some kind of price / performance optimum: The Korad KEL2010 and Tonghui TH8402A (also sold by BK precession as BK8550 / BK8551 and perhaps other models). Now the KEL2010 isn't the most compact unit but it does a reasonably big display and it has front to back cooling so it can be stacked tightly between other equipment.
See
https://www.eevblog.com/forum/testgear/dc-load-korad-kel2010-or-tonghui-th8402a/Long story short; I couldn't get the TH8402A / BK855x quickly from a seller that offers hassle free returns so I ordered a KEL2010 unit from Eleshop to try it. The KEL2010 is priced at 434 euro (ex VAT). As usual Eleshop's service is excellent as they where able to provide me with the missing Korad manuals within a couple of days (and put these on their website for everyone to access).
As I intend to use the DC load for relatively low currents as well, I bought the lowest power, lowest current model. I already have a high power DC load.
The manual I downloaded from Eleshop is slightly confusing as the PDF file has the pages ordered for printing the manual as a booklet. Fortunately, the KEL2010 comes with a printed version of the manual.
Firmware: 1.14 / comm: V1.12
Test planItems to test:
- User interface
- Warm up behaviour
- Regulation stability at 1mA, 5mA, 10mA, 50mA, 100mA, 500mA, 1A, 5A
- Regulation resolution
- Regulation ripple
- Dynamic behaviour (pulse test)
- with battery
- with linear PSU
- Constant voltage control mode
- On screen plotting
User interfaceThe user interface is OK. The display is easy to read with voltage, current and power which are prominently displayed.
When you key in a new number for the setting, the output is turned off. This prevents making mistakes.
One of the things high on my list are binding posts that can also take 4mm banana plugs and the KEL2010 has these.
The graph on the unit can be scaled & controlled using the up, down, left and right buttons. But somehow the setpoint (current for example) can not be adjusted after changing the graph. It takes changing modes a couple of times to get setting to work again. Definitely a bug.
Warm up behaviourWarm up behaviour looks good. After warming up, the level stabilises. I used a DC current of 100mA to test.
Regulation stabilityFor these tests I used a Li-ion battery to make sure there is no interaction between the control loop of a power supply and the DC load. For the current logging I used my Keysight 34461A 6.5 digit DMM.
I tested at various currents and it looks like the current is about 1mA low on my unit.
1mA:
10mA:
When testing at 500mA, the fan switches on every few minutes. The temperature variation is clearly visible in the graph but the peak-peak deviation is about 0.5mA so not bad. But it does seem like the fan is introducing more noise into the current signal (rising edges with the fan on are fuzzy).
Regulation resolutionThe current can be set in 100uA increments but the actual steps seem to be 600uA to 700uA in size which corresponds roughly with 45A / 65555 (16 bit resolution). Thus giving a bit of over overrange on the current measurement.
Regulation rippleThe DC measurements show promising numbers but how about the AC ripple?
For this measurement I used a 10 milli-Ohm shunt resistor and my Yokogawa DL708 oscilloscope (with a floating input module that has 12 bit resolution). This oscilloscope is designed to measure low level, low frequency signals. I set the bandwidth to 5kHz and the vertical V/div is scaled to 10mA per division (in reality this is 100uV per division!). Now there is a different picture that shows some ripple. With the KEL2010 set to 5A and loading the battery, there is a 50Hz wave visible which is around 15mA peak-peak. When the current is set to 100mA, the ripple remains the same. The blue trace is the frequency spectrum.
When the fan comes on, things get worse. There is a large amount of noise riding on the load current. The ripple is ballpark 20mA peak-peak.
BTW, the reason the 50Hz ripple doesn't show up on the DMM is because the DMM samples synchronously with mains so any mains noise is cancelled by the measurement. The fan shows up barely because the frequencies from the fan are random to the DMM which only measures DC average current.
There is also a difference between low speed and high speed control mode. Oddly enough, the low speed control mode has overshoot while the high speed control mode seems to behave much better. Likely some loop control parameter has not been tuned correctly in the signal processing. I'm still waiting for Eleshop / Korad to come back to me on this whether this is a bug or somehow intended behaviours for certain loads. However I'm inclined to think the slow control loop behaviour is a bug because a Li-ion battery is close to an ideal voltage source and no current control loop should have problems with an ideal voltage source. I made all the measurements using the high speed control setting.
Dynamic modeThe dynamic mode is a bit unlogical. You get tables with values. To change a value, you have to select 'save dyn' and use the enter key to enable a field for editing and then press enter again after entering the value. To activate the new values, you have to use the 'call dyn' button followed by enter and go back to the main screen before you can use the dynamic values. This could have been made so much easier. But I guess the designers really wanted to seperate editing mode from operational mode and make it hard to change settings on the fly. I can see why this is done to prevent mistakes though.
Some images taken with the slow CC mode:
Some images taken with the fast CC mode:
Fastest setting:
10mA per us
All in all the load behaves well both with a battery or a power supply in high speed control mode. At really low voltages (below 3V) ringing starts to occur.
I tried settings up to 30A (at 8V) and behaviour seems to be OK.
Constant voltage / constant resistanceI tried both these modes on a current sourcing LED driver. The constant voltage mode didn't work (pulled the voltage way too low) in both slow and fast control loop modes. The constant resistance mode did work.
Fan noiseAs a relatively low cost DC load needs to convert all power into heat, there is no way to avoid a fan (especially in a compact unit). The fan sounds raw and you can feel the outside of the case vibrating while the fan is running. In other words: the fan used is not the highest quality. A better quality fan suspended in rubber mounts is likely a good upgrade.
Fortunately the fan is temperature controlled. When the load is light (say a couple of Watts), the fan is off. At high load the fan speed is controlled up/down in steps so it is not at full blast when it is not needed.
Remote controlFor remote control there is RS232, USB and LAN. The LAN communication is a bit of an oddbal because it uses UDP packets to convey SCPI commands. A simple piece of Python code should take care of it but it would have been nicer if I could use TCP/IP based Pyvisa to deal with communication. I have not tested whether remote control works.
A look insideAt first glance the case and electronics look well build. I have a 300W unit but it looks like it has the driver circuits for the 500W model. Only things missing are the current sensing resistors and MOSFETs. It also looks like the board has plugs to extend. So likely the full width (19") units have an extra load board with two more heatsink + MOSFET sections.
The front panel & logic is driven by an NXP LPC1788. Kind of surprising to find what some would consider a premium part in here. But maybe this is one of the very few options that can drive a TFT screen directly.
Pick & place mishap:
All the analog electronics and power sits on 1 bird board that spans the entire bottom of the KEL2010. It looks decent but the electrolytic capacitors could have been placed further away from the heatsinks.
The ADC is a 16bit, 250ks/s type from AD: AD7682. An AD8656 dual, low noise opamp seems to be used as a pre-amplifier for the ADC.
The I/O (RS232, USB and ethernet) is on an isolated section of the board. There is a mystery microcontroller (the USB driver mentions Nuvoton) that seems to take care of the RS232 and USB directly. Ethernet is handled by a Wiznet chip. With the Wiznet chip in there, it is odd that the KEL2010 doesn't support remote control over TCP/IP. Handling TCP/IP is just as easy as UDP. The LPC1788 also has a MAC on board but an external PHY costs as much as the Wiznet chip. On top of that the Wiznet chip is much easier to use for relatively simple ethernet communication tasks so it lowers the software NRE costs substantially.
A picture of the solder side of the board: This looks good; at least some thought has been put into the layout. The analog part seems to have an almost solid ground coverage.
An interesting detail is that the DC load section itself is fully isolated from ground and the I/O interfaces at the back. However the ground of the front panel is shared with the ground of the DC load. Which also means that the ground of the USB connector on the front is connected to the negative input of the DC load. This is something to be aware off when using the DC load at elevated voltages, a USB stick with metal parts may not be safe to touch.
From the component choices it is clear the designers made concise choices where it comes to part cost versus software development costs. As mentioned above, the use of a Wiznet chip instead of a software stack is a clear indication.
Still, it is amazing they can retail this unit for 434 euro (ex. VAT). It can't cost more than 100 euro to manufacture otherwise there is no way to make a profit.
Fixing the rippleIs it worthwhile to attempt this? I think so. Without the 50Hz and fan current ripple, the KEL2010 is capable of single digit milli-amp current regulation. The electronics seem to be designed carefully enough but something simple may have been overlooked.
I took some measurements using the DL708 (again) at low bandwidth and 100uV/div using a 1:1 probe. There is some residual mains (50Hz/100Hz) noise on the supply lines (in the 300uV pp ballpark) but the opamps should have a high PSRR in that frequency band. But somehow power supply noise find its way into the circuitry...
First a bit of analysis of the circuitry:
The load section is build classically with 1 opamp per MOSFET. The opamps used are TL082 dual opamps from ST. U19 seems to be the buffer that feeds the load sections through R7. In turn U19 is fed by U9 (DG211B analog switch) through some circuitry (R47, Q10 and Q12). It looks like Q10 & Q12 are controlled by the front panel.
U15 (OP07 single opamp) is a buffer amp that feeds U9 (DG112B switch). In turn U15 is fed by U3 (another OP07).
U30 looks like to be the DAC that drives the output signal. It is marked with DH7 and an analog devices logo. The closest match that makes sense with the circuit I could is the AD5693 series I2C DAC.
First I want to check the noise on the DAC output (R6), reference voltage (R42) and the signals around R7, R47, Q10 and Q12. This looks clean. When I check R7, I can see a 50Hz signal appearing when the DC load is switched on. So the 50Hz noise is coming from somewhere upstream. Back tracing to U15 shows the same behaviour. This opamp circuit is fed by R60 and R76. R60 show opposite behaviour (50Hz when load switched off). U15 and U19 form some kind of feedback loop with U9 switching it on and off. When on, the 50Hz current can flow into R76 making it dissapear at R76. This is not making things much clearer. I start to think this is some kind of ground related issue.
To find out what is what, I tried to add more capacitance to the output of the rectifiers to see if the issue is not somehow related to the line regulation of the regulators (a bunch of good old 78xx and 79xx to make primary voltages like 5V, 12V, -15V and +15V). It quickly turned out adding more capacitance didn't make a difference though. So the problem is not in the electronics itself but the noise is being picked up somehow (induced into the circuit).
Next thing I tried was taking the transformer out of the chassis and BINGO! The further the transformer is away, the better it gets. So even though the transformer is a shielded type, it still puts out a strong enough 50Hz field to cause problems. Now onto the fan causing current ripple. Would that be magnetic induction as well? Yes, it is! When powering the fan from an external power supply, the noise in the current starts to show up as well.
I'm most annoyed by the 50Hz ripple so I decided to mount the transformer onto the rear plate. This is a simple fix that only requires making a short extension cable. The rear plate is sturdy enough to hold the transformer mounted on a couple of 45mm M4 standoffs. Secondly I swapped the fan for a higher quality model from Sanyo Denki ( 9GA0812P4H001 ).
As an extra measure I made a shield out of a piece of 0.4mm mu-metal I salvaged from a UPS a long time ago. With the shield, there is already a 30% reduction of the 50Hz ripple when the transformer is mounted in the original position.
The result of these simple fixes is excellent. The 50Hz ripple is reduced to around 1mAp-p. I had to use averaging to make the signal visible. But that is also due to the measurement setup I used; the oscilloscope is set to 50uV/div here. I'm measuring single digit milli-amps using a 10milli-Ohm shunt resistor. The current ripple introduced by the is fan hardly visible so that is a major improvement as well compared to the original fan.
ConclusionAll in all this a pretty good DC load where it comes to accuracy, stability and regulation resolution (at least using the high-speed control setting). I've found some good quality chips in there as well.
But it needs some modifications to step it up. If Korad ever decides to do a re-design, they should move the transformer and the fan to the back of the unit. As far away as possible from the sensitive circuitry. Maybe a metal shield over the electronics also helps to improve things further.
The user interface can use a bit more polishing. Some of it is just not logical; it is a typical UI made by a software engineer. It is another showcase that the major cost of modern test equipment is in software engineering and not so much in the cost of the hardware.
Meanwhile, the KEL2010 has come in handy for a couple of projects already.
