"Digital" power supplies

[Phaedrus]

It's my understanding that a Digital SMPS means a PSU where the analog op-amp based feedback circuit is replaced by an ADC and microcontroller running an algorithm. The point being that you can reduce the effects from variations in analog components and can tweak your algorithm specifically for your topology. You can also connect it to a computer and get real-time data on your PSU's state and performance.

What some companies are doing in the ATX business is taking an analog PSU, and sticking a USB controller and a comparator on it to give you output voltage and fan speed readings. Example:

http://www.hardwareluxx.de/index.php/artikel/hardware/netzteile/28924-roundup750wdez2013.html?start=3


That's using an industry standard CM6901 resonant controller. All they did was put a Ti USB controller on there, and claiming it's digital. Having worked on an actual DSMPS before (EVGA NEX1500; not the best PSU ever made for a number of reasons, but it had real technical grunt behind it...) this kind of pisses me off.

What's your opinion? Is a USB controller enough to make a PSU "digital"?

[Mr Smiley]

What about the Dyson 'Digital' Motor 

And now that most of the 'Analog' TV channels have been turned off, there transmitting in 'Digital' 

There are good points, c4757p is as happy as pigs in .....     he can only work with digital things, he thinks a True Analogue LCR meter isn't a proper LCR meter cos it doesn't have a lcd display and a test button. 

 

[AG6QR]

It's my understanding that a Digital SMPS means a PSU where the analog op-amp based feedback circuit is replaced by an ADC and microcontroller running an algorithm.

If that's what it means, consider that the transient response would have to be delayed by the time required to read the ADC, plus at least a few instruction cycles to process the information, and most likely the time to write a control parameter to the DAC.  That could be fast enough for many purposes, so it may not matter, but it's probably not as fast as a well-done analog op-amp feedback loop.

If you need fast transient response, it may be nicer for the feedback loop to be done in fast analog, with the voltage (and current if applicable) setpoints controlled digitally.  See the long series on Dave's microsupply for one example of how to control a supply digitally.

Regardless, I don't trust the marketing people enough to ascribe much of any particular meaning to the term "digital".

[SeanB]

And they even short change you with caps as well, just putting in enough to barely get by even though the footprints are there for the extras to be a lot better than minimum spec. At least there is a chance the reading will be close to correct, though with the thin cable they are using what the motherboard will get is anybody's guess, it will for sure be not the voltage at the supply side.

[G7PSK]

What about the Dyson 'Digital' Motor 

And now that most of the 'Analog' TV channels have been turned off, there transmitting in 'Digital' 

There are good points, c4757p is as happy as pigs in .....     he can only work with digital things, he thinks a True Analogue LCR meter isn't a proper LCR meter cos it doesn't have a lcd display and a test button. 

 

A few years ago the "wank word" of the day was "turbo" now its "Digital" who knows what it will be in a few years time turbo digital or digital turbo but rest assured the marketing gurus will have something that they think trendy.

[Andy Watson]

A few years ago the "wank word" of the day was "turbo" now its "Digital" who knows what it will be in a few years time turbo digital or digital turbo but rest assured the marketing gurus will have something that they think trendy.

A digital turbo-encabulator ! Now I could really use one of those.

[free_electron]

pff. digital is just a heavily constrained version of analog.

i wouldnt trust a digital regulation loop for a heavy duty power supply as far as i can throw it.

[Neilm]

pff. digital is just a heavily constrained version of analog.

i wouldnt trust a digital regulation loop for a heavy duty power supply as far as i can throw it.

I looked using at one of those a year ago. I decided then that you could get one that would work similarly to to an analogue system. Trouble was the controller required was expensive, as was the fast ADC required (10M Samples/sec). The main reason I wanted it was thinking I would have to adapt the control loop mid test. After some testing, I decided I could do all the control I needed in the analogue domain with a different control chip.

[Phaedrus]

In the ATX PSU market, few reviewers have the equipment to do transient load testing, so the SOP is to design it to have "good enough" transient response, then focus on achieving the best possible steady state regulation and ripple, which is what the top reviewers score on. In any case, the heavy transient loads are powered by the CPU and GPU buck regulators, and their input capacitance slows the slew rate on the PSU, so transient performance isn't all that critical anyway, unless you're overclocking on liquid nitrogen.

[Christopher]

i work for a power supply company in the as a junior engineer.

We mainly use trusty(?) unitrode controllers for the PFC/Boost (high voltage for holdup caps) and housekeeping stages in proven blocks. Output bucks/flybacks from the boostare normally custom as well as the analog power good and synchronization.

We have a new ex RF engineer who is working on AC(115v)/DC to isolated AC(110V) and has implemented a digital stage. It works, but the manager doesn't trust digital as far as he can throw it as free said ...

I have been experimenting with PICs for the control loop but it's no good. Microchip bought out a PIC which has a seperate analog stage for PWM control and you can set the reference voltage with a register which is neat, but no better than using a PWM controller and a DAC on the feedback, infact I think it would limit you..

[mariush]

There are power supplies that go a bit more than just a usb link to report voltages and stuff, like the corsair axi power supplies (flextronics oem ) :

http://www.jonnyguru.com/modules.php?name=NDReviews&op=Story5&reid=300

Here we have the DSP Control Board. Over on the right side we have a MC56F8014 VFAE 16-bit Digital Signal Controller. In the lower middle we have a C8051F380 USB Flash MCU, and just above that we have a C8051F310 32 Pin Mixed Signal MCU.

[Phaedrus]

Yes. That's a real DSMPS. The Corsair AXi series and EVGA NEX1500 (which I worked on) are the only digital ATX PSUs on the market. The Thermaltake I linked is newly announced, and is basically riding on the coattails without doing any work. There's not even really excuse, since Thermaltake's main OEM is Channel Well, and Channel Well has a digital solution almost ready for deployment...

The real problem is that using a digital controller is an extra ~$30 BOM cost plus NRE charges and programming costs. All for a technology not yet proven to drive sales outside of the Corsair AXi series, which mainly sells because Corsair is the Apple of the PC enthusiast world.

[Christopher]

The way these PC PSU reviewers review pc power supplies is a little worrying, don't you think? Details please, Is ripple measured at max load? PARD filter?

I even saw one who measured the output voltage from the e-load rather than at the output connector..

[c4757p]

Pardon the curmudgeonish reaction here... but why do we even need to "innovate" with freaking power supplies? They take power from A, regulate it and dump it to B! How hard is that!? The only reason I can see to ever bother redesigning an ATX power supply is to make room for a different size part or to rearrange for the new controller chip that's cheaper to produce now than when you originally designed it...

Analog feedback loops work. Don't the engineers have something better to do?

[NiHaoMike]

There's a trick you can play with PFC to make it a little more efficient. Basically, PFC is a boost converter, and as such, the instantaneous efficiency goes down as the instantaneous input voltage decreases. With digital control, it's possible to adjust the cut in point to boost efficiency, while still getting a good power factor. The savings can be especially pronounced at medium load (i.e. where it normally operates at!).

Both the main converter and the PFC can boost efficiency with period skipping (or an equivalent) at low loads, though analog can also do that.

What I would like to see is an ATX PSU with a built in UPS, with the batteries either connected externally or mounted in drive bays. My homemade PC actually has something like that. I opted for a separate battery charger (rather than having the main converter operate as a charger) and there's a USB link that can do all sorts of neat tricks like monitor real time power use and even change some operating parameters.

[Phaedrus]

PC power supply reviews vary pretty wildly in terms of testing, equipment, and knowledge. There's only a couple guys who do transient loads; one has a resistor bank and some high current relays. Another uses hard drive motors (which is a problem since that's a very inductive load). There's one place that does very nice cross loading tests where they do a 5 second test at every 1Ax1A combination of +12V vs. +5V/+3.3V, and dumps it into a very nice graph. Everyplace else does step loading (always 20/50/100% load. Occasionally others). Most sites that know enough to use E-loads also know enough to measure voltage at the connector, not the loader. PARD is done to Intel spec, with a 10uF and 0.1uF cap very close to the probe point.

Almost always tested for:
Absolute voltage regulation (within +/- 5% of nominal value)
Relative voltage regulation (minimum voltage divided by maximum voltage)
Ripple & Noise (at full load at least)
Efficiency (focus on achieving 80PLUS levels)

Often tested for:
Overload test (110%, 120% and/or up to maximum PSU can provide)
Short circuit tests
High temperature performance (40 or 50 degrees)

Occasionally tested for:
Transient response
Comprehensive cross-loading
Hold-up time
Sound testing

Never tested for:
Conductive/Radiative EMI

[tesla500]

pff. digital is just a heavily constrained version of analog.

i wouldnt trust a digital regulation loop for a heavy duty power supply as far as i can throw it.

Better not get that Tesla in that case, 400kW under control of a little micro...

To be fair, it would be very hard to do vector control in analog, as opposed to a simple power supply feedback loop.

[codeboy2k]

That Corsair AX1200i is sure built well, it's not your run-of-the mill typical PC PSU, that's for sure, but $350 

I see how they have put the 3.3V rails, 5VSB rails on risers, they can probably build different versions using different risers.

Pardon the curmudgeonish reaction here... but why do we even need to "innovate" with freaking power supplies?
[...removed...]
Don't the engineers have something better to do?

They need to innovate in order to drive sales.  Features drive sales.  They need to keep adding features or else there's no reason to buy a new PSU, or the best PSU, top of the line, etc... There's always going to be some kiddy who wants this, and is excited by being able to chart their 12V  and 5V rails over time...

Future features will likely include battery inputs (you can run the PSU directly off two 12V SLA batteries, just insert them into the circuit after the boost stage. During a power failure the DC bus drops from 400VDC to 24VDC --- as long as the system is designed for the two different DC bus voltages, everything will be fine)

[Phaedrus]

With a DC-DC PSU (+5V and +3.3V buck regulated from +12V) you can just put the batteries after the +12V rectifiers and that would work as well.

[PedroDaGr8]

PC power supply reviews vary pretty wildly in terms of testing, equipment, and knowledge. There's only a couple guys who do transient loads; one has a resistor bank and some high current relays. Another uses hard drive motors (which is a problem since that's a very inductive load). There's one place that does very nice cross loading tests where they do a 5 second test at every 1Ax1A combination of +12V vs. +5V/+3.3V, and dumps it into a very nice graph. Everyplace else does step loading (always 20/50/100% load. Occasionally others). Most sites that know enough to use E-loads also know enough to measure voltage at the connector, not the loader. PARD is done to Intel spec, with a 10uF and 0.1uF cap very close to the probe point.

Almost always tested for:
Absolute voltage regulation (within +/- 5% of nominal value)
Relative voltage regulation (minimum voltage divided by maximum voltage)
Ripple & Noise (at full load at least)
Efficiency (focus on achieving 80PLUS levels)

Often tested for:
Overload test (110%, 120% and/or up to maximum PSU can provide)
Short circuit tests
High temperature performance (40 or 50 degrees)

Occasionally tested for:
Transient response
Comprehensive cross-loading
Hold-up time
Sound testing

Never tested for:
Conductive/Radiative EMI

The two main ones I follow are JG and [H]. The later took a lot if advice from JG as he was a frequent member there as well as on his own site. As such the testing methodologies are very similar.

Sent from my EVO using Tapatalk

[codeboy2k]

I like this site's  PC PSU testing setup and methods [Tech PowerUp]

TechPowerUp PSU Review Methodology (4 page article, lots of pictures)
http://www.techpowerup.com/articles//overclocking/psu/161

TPU PSU Performance Rating Methodology Explained (2 page article)
http://www.techpowerup.com/articles/overclocking/psu/162