Quick look at Agilent E3620A low-cost 2 channel DC power supply


In-house miniTEC chamber requires two power rails to properly operate. One rail is fix +12V to power the hotside TEC cooling setup with waterblock and fansink, while second rail adjusted down +7 V … +8 V for the mixing fan inside of the airbath for better uniformity and operation with active DUTs.


I’ve used Vicor AC-DC dual channel module before but it died few months back and question with new power source arise. Going back to our favorite second hand used equipment source one of older Agilent E3620A Power supplies was acquired for reasonable $60 USD. E3620A power supply has nice set of features:

  • Output 1: 0 to 25 V, 0 to 1 A
  • Output 2: 0 to 25 V, 0 to 1 A
  • Total output power (max): 50 W
  • Ripple & noise from 20 Hz to 20 MHz : Normal Mode Voltage rms: 350 ┬ÁV, Peak-to-Peak: 1.5 mV
  • Load & line regulation : 0.01% + 2 mV
  • Readout meter resolution : Voltage: 10 mV (0-20 V), 100 mV, (>20 V), Current: 1 mA

This is a simple and cost-optimized (1154 USD brand new at Keysight as of December 2023) 50W laboratory supply, with two isolated channels. There is also triple channel version of this supply , model number E3630A.

Supply is fanless and built around linear regulators and massive transformer. E3620A does not have remote control or interface capability and always provides outputs preset by nice quality multi-turn potentiometers at the front panel. This is perfect feature for the supply with miniTEC chamber which require power to keep integrated Peltier element within nominal temperatures. If mains power would be cycled power supply would not reset like with modern digital-based supplies and will resume power output automatically without any special programming/settings. This power supply does NOT support sinking current.

Power supply has massive die cast heatsink at the back to dissipate heat from active pass transistors used in each channel. Because it is not switch-mode supply efficiency would be relatively bad, especially with lower output set voltages. There is no rear terminals for power output. Only front terminals are available to the user. E3620A does not have remote sense input either but it is not a significant issue given relatively low maximum 1 A current limit that these supplies offer.

Negative 5-way post for second power supply channel is missing plastic shroud. But it still can be used normally using metal nut. I might replace it in the future if I can find suitable binding post that would fit.

After removing plastic cover held by latches we can see the internal design. PSU is built on a large board. Full schematics is available in older E3620A service manual, which is extremely useful for troubleshooting and repairs. There are no fancy processors or firmware in this design, making it also more repair friendly. Sadly Keysight removed all of the schematics from the latest E3620A manual, following with unfortunate anti-repair industry trend.

Internally this unit is pretty clean with only few spider webs and one dead spider inside. Making a home in a closed up power supply was not the best survival strategy on a spider’s side.

High power components equipped with aluminum heatsinks and main output pass transistors in TO-3 package are cooled by massive die-cast heatsink with fins at the back of the chassis. Bottom of the PCB is not accessible without taking the front panel off, but I don’t expect anything too exciting there.

Right side of the front panel plastic bezel had cracked features and is a bit loose. It is pretty simple fix with some superglue to fill the cracks and keep everything secure. Front panel design is quite typical for HPAK equipment in this form-factor even today.

Linear low-frequency power transformer is mounted to the PCB with massive metal studs and sealed in thick protective lacquer. There is no audible hum or noise from this power supply under operation.

Output is adjusted by very nice multi-turn wire-wound potentiometer. Each channel is adjusted by own potentiometer and digital 3½-digit voltage/current meter is selected to display either channel by two switches V1 and V2 under the display.

Power supply built with good quality components, top tier electrolytic capacitors from Nichicon and Chemi-con and implements thru-hole design layout. No repairs were necessary to circuits in this unit. According to datecode on chips E3620A was built in the end of 1999.

Calibration of the power supply is done with DC load such as recommended 0.1 Ω 3W resistor and accurate DMM, such as HP/Agilent 34401A. I had Keysight 3458A idling nearby, so it was pretty easy selection for a quick test :)

Without any adjustments in condition as received power supply own meter was just about +0.033 V off and voltage output is nice and stable. After adjustment of the meter readout with blue trimpot located at the front display board it was easy to adjust supply with a millivolt from 3458A readout. Second trimpot is responsible for current readout channel and calibrated with Keithley 2425 SMU sinking 1.000 A.

Over-current limit protection in this E3620A is set at 1.06A. Well enough for intended application at xDevs lab.

Schematics of channel 1 output circuit is presented below. Design of second channel is identical and using separate transformer tap to maintain isolation.

Efficiency measurement

To test efficiency E3620A under test was powered from Chroma 61604 programmable AC source with integrated analysis functionality. Output of the E3620A connected to programmable Chroma 6334 DC load. All data was measured with AC sourced programmed at fixed 120V 60Hz output.

With nothing connected to the PSU outputs power draw from AC was around 7.7W with both channels set to 0V and 9.8W with both at maximum 27.5V. At 0.5A @ 10V loading for channel 1 only power draw increased to 31.2W. With one channel loaded to maximum 1A at 25V power draw increased to 53.7W. Both channels loaded to same power brought total AC input power to 100.8 W. Basic calculations can be used now to determine efficiency figures.

This is well in line of expected from a simple linear power supply. PF is around 0.6 at the full load. Same data set also available in table form:

V1 output, V V2 output, V I1, A I2, A Pout DC, W Pin AC, W Efficiency
0 0 0 0 0 7.7
27.6 27.5 0 0 0 9.8
2 0 0.2 0 0.4 16.6 2.41%
4 0 0.2 0 0.8 16.7 4.79%
10 0 0.2 0 2 17.5 11.43%
10 0 0.5 0 5 30.5 16.39%
20 0 0.5 0 10 31.2 32.05%
25 0 0.5 0 12.5 31.2 40.06%
25 0 1 0 25 53.7 46.55%
25 20 1 0.2 29 62.3 46.55%
25 25 1 1 50 100.8 49.60%
27.6 27.4 1 1 55 100.9 54.51%

Noise measurement

To test noise manufacturer recommends loading power supply output with 25 Ω 100W 5% resistor and connecting true RMS voltmeter. Expected noise value should be under 350 µV RMS. Noise is measured with help of Tektronix ADA400A differential amplifier configured with gain 100x. Output of the amplifier connected to Tektronix DPO7104C oscilloscope, limited to 20 MHz bandwidth. Amplifier was configured with differential AC-coupled measurement.

Noise floor with grounded inputs was just 50 µV peak to peak, well below expected signal level from the power supply. Next step is to connect 25 Ω high-power resistor to load up supply and configure it for 15VDC output. Only channel 1 was actually tested, I expect second channel to follow same performance.

Measurement with 100 kHz bandwidth limit at ADA400A provided us with 905 µV peak to peak or 305 µV RMS sine with frequency very close to 60 Hz, which is just AC ripple from linear regulator.

Limiting bandwidth to 100 Hz gives slightly smaller noise numbers at 801 µV peak to peak or 271 µV RMS. Well within expected E3620A specifications.

Stay tuned and let us know your feedback on this post! Discussion about this and related stuff is also welcome in comment section or at our own IRC chat server: xdevs.com (port six-zero-ten-zero, channel: #xDevs.com) or via e-mail.

Author: Ilya Tsemenko
Created: Dec. 10, 2023, 5:32 a.m.
Modified: Dec. 11, 2023, 2:58 a.m.