LRS 001: Regarding PC PSU functionality
- Section 1: Reliability
The unit must be designed for a service life of at least five years (43,830 hours) for the main supply, while delivering 80% of the rated power with 45°C at the intake, and ten years (87,660 hours) for the standby supply. At the end of this service life, the output ripple must remain within the allowed limits (120mV for +12V and −12V and 50mV for the lower voltages) at all power levels. (I realise that meeting this standard will pretty much rule out 85°C electrolytics even for the primary cap.)
The unit must be able to deliver full rated power with 50°C at the intake (this is based on 40°C at the PC case's intake, plus a 10°C rise from there to the PSU), at the lowest input voltage. The unit is not required to last for the full service life under those conditions, but if it can be made to do so, that's great. - Section 2: Output cabling
Along with being of appropriate lengths, the output cables should have appropriate conductor sizes for the current draw and acceptable voltage drop.
For the CPU and PCIe power cables, the total drop should be below 0.12V (1% of +12V) at the official load specification (for PCIe connectors) or 5A per wire (for the CPU connectors, unless the unit imposes a lower limit).
Power cables for disk drives should be sized for a total drop below 0.24V (2% of +12V) with a worst-case spin-up current of 2.5A per drive. (In some cases, staggered spin-up may be needed to avoid excessive drops on cables shared between multiple drives.)
For other cables, a reasonable maximum load should be used with a total drop of 0.12V for +12V, or 0.1V for the lower outputs. (Sharing of grounds between multiple outputs may complicate the maths.)
In all cases, the wire gauge must be adequate to not exceed the rated maximum temperature of the insulation at an ambient temperature of 50°C. For low-current wires (−12V, Power Good, On/Off and the sense lines) the smallest gauge suitable may be determined by compatibility with the crimp terminals, rather than losses.
The labelling on these is presently quite inadequate, with no indication of the ripple or regulation. Any mention of efficiency is limited to just a few general "classes". So:
- Section 1: Labelling
First of all, supplies designed only for the low range (100–127V~ nominal) must have a very prominent warning (on the unit itself, not in the manual which most people will never read) that they will be destroyed if connected to the high range (200–240V~ nom.). Supplies with a manual voltage selector (which are admittedly rare on external PSUs) must have an equally prominent warning to set it correctly.
In addition to the nominal input voltage and frequency, maximum inrush current should be specified, if more than 1A or double the operating current (whichever is greater). Efficiency and power factor under full load should also be specified, as should standby (no load) draw and power factor.
Along with nominal output voltages, the regulation (including the output lead's resistance) and (with the obvious exception of AC-output linear supplies) ripple (allowing for capacitor aging) should be specified. Switching supplies must have a complete list of protections (more on them below). - Section 2: Design and functionality
If the unit cannot be made compact enough to fit alongside other plugs of the type used in the region the unit is sold in (or if it wouldn't be economically viable to make a customised version for a small market), it should use an inlet from the IEC 60320 series. The mounting of this inlet must be rigid enough to withstand at least 1,000 plug/unplug cycles without failure. If the standby draw is greater than 1W (as with most linear supplies) or 2VA, I recommend including a switch on the unit itself. This switch should be rated to last for at least 10,000 switching cycles with the voltage and current (including inrush) applied.
Due to the risk of strong static discharges, Y capacitors should not be used with Class II. Larger transformers may themselves have significant primary-to-secondary capacitance, which must be assessed.
The lifespan must be at least ten years at 80% or lower load, or five years at full load, with an ambient temperature of 40°C. Linear supplies will likely last much longer (especially the AC-output types, which have no internal components besides the transformer itself).
Switching supplies must have OVP and SCP, and OCP is also recommended. OTP is required for fan-cooled units; for passively cooled types it is optional, but if not implemented, a DO NOT COVER warning is then compulsory. If a fan is to be used, it should be temperature controlled to reduce noise and dust accumulation and prolong the lifespan of the bearing(s). 2BB fans (although allowed) may be a poor choice for this application due to limited shock resistance, and 1B+S configurations are prohibited; FDBs or similar are preferred, but ordinary sleeve bearings may be good enough for less demanding applications. A dust filter (that can be cleaned by the user) should be used.
The output lead should be of a suitable length (I suggest 2m) for general usage, and I suggest calculating the conductor size for a drop of 1–2% of the output voltage (or a conductor temperature below 60°C at 40°C ambient, whichever is the primary issue). (Some chargers for cell phones or portable game systems use a lead with much higher resistance, and the device is designed to tolerate the poor regulation.)
I recommend that the casing be held together with Torx screws, or failing that, plain old Phillips types.
And one last thing – no half-wave primary rectifiers!
On a semi-related topic, I was wondering about the feasibility of an external SMPS with a solid aluminium case (anodised black for maximum radiative cooling) acting as one large heatsink. Better cooling than the standard plastic-cased type with no fan noise – presumably.