PSU Linear vs Switch-mode front end?

[obuone]

Hi guys and galls
I'm a beginner in field of electronic engineering and as a starter I want to do a PSU design.  I want to incorporate the advantages of both linear and switch-mode architectures by combining the small size and high efficiency of the SMPSU and the linear output of an LPSU. The switching part is going to track the output with a few volts above the dropout voltage of the liner controller.

So my question is how is the linear controller going to be affected by the switching crap and is it going to capable of regulating the output voltage properly?

[Kleinstein]

When doing simulations its usually a good idea to treat the SMPS plus filter and linear regulator as two separate parts. So if noise / HF ripple is important you should have an SMPS that does not have too much ripple and a linear regulator stage with good rejection. At least for the very high frequencies (e.g parasitic ringing) a LC filter is usually more efficient than the linear stage. So one usually has an extra filter stage in between.

For a beginner I would start with a separate SMPS and linear regulator. Putting them together is than only a small step.

[dannyf]

good news and bad news.

good news: it works. I have built multiple modules and without exception they all work and are particularly useful in cases of wide output range (large Vin - Vout), and high current load - I typically design for 3 - 4 voltage drop over the linear regulators.

bad news: it is not terribly effective for ripple rejection. datasheets typically quote a ripple rejection of 60 - 67db, at 100/120Hz however. Your typical smps works at 50  - 150Khz and at that frequency, your psrr can be as low as 20 - 30db. For effciency and size, you want high switching frequencies; for psrr, you want low frequencies.

So I concluded that for analog uses, it is not worth it unless 1) you use integrated linear regulators; and 2) you can find a low switching frequency / low ripple smps.

For digital uses, go with smps directly.

[nctnico]

bad news: it is not terribly effective for ripple rejection. datasheets typically quote a ripple rejection of 60 - 67db, at 100/120Hz however. Your typical smps works at 50  - 150Khz and at that frequency, your psrr can be as low as 20 - 30db. For effciency and size, you want high switching frequencies; for psrr, you want low frequencies.

Hint: filters

[Siwastaja]

bad news: it is not terribly effective for ripple rejection. datasheets typically quote a ripple rejection of 60 - 67db, at 100/120Hz however. Your typical smps works at 50  - 150Khz and at that frequency, your psrr can be as low as 20 - 30db.

More bad news: switching frequency is rather irrelevant. Your typical SMPS emits switching noise into mega-gigahertz range; it's from the slew rate of the square wave. And at a few dozen megahertz, your linear reg PSRR is even lower.

So, you don't even try to get rid of switching noise by using a linear reg post-stage. That's not the point at all. Well, what's the point then? I guess one of the points is to increase transient response, for load steps, for example. And yes, linear stage helps with mid-frequency noise somewhat.

SMPS-linear hybrid is not a silver bullet. Proper SMPS design is paramount; after that, the filters. Adding a linear stage comes long after, IMO.

So, if you are going to design your own SMPS, don't even think about post-regulation steps at this point. It will draw your focus to irrelevant control questions. Dedicate your time on learning how to design an SMPS, because that's where it starts. After you have a proper SMPS going on, the odds are that you are more than happy with it as-is, and probably designing an extra LC output filter stage comes next.

[Kalvin]

Linear technology has some ultra-low noise switching mode regulators. For example, the LT1533 has typically 100uV p-p noise ripple voltage. Other manufacturers may have similar products.

(The LT's site seems to be missing the data sheets at the moment. Maybe it is going through some maintenance or some server is down)

[Kleinstein]

The LT1533 won't help much, as this is a low power transformer driver. More to build a 1-5 W fixed output DCDC converter.

Usually lower frequency SMPS can have less noise in the high frequency range to, as they can use slower switching without to much penalty on efficiency. Very high efficiency circuits tend to have more noise.

Having the linear regulator to compensate for a voltage drop and dynamic performance there can be quite some filtering between SMPS and the linear stage. With highly dynamic loads this might need some extra voltage drop for the linear stage.

[Kalvin]

I just used LT1533 as an example of a low noise switching regulator. There may be some other more appropriate devices available from Linear and possibly from other manufacturers. The "secret" of low noise switching regulator is controlled signal slew-rates ie. rise and fall times, though.

[Siwastaja]

Having the linear regulator to compensate for a voltage drop and dynamic performance there can be quite some filtering between SMPS and the linear stage. With highly dynamic loads this might need some extra voltage drop for the linear stage.

Yes, I think this is the point.

The linear stage is not the noise-eliminating step per se.

The filtration must be done separately, but as the filtration makes it even more difficult to achieve good transient response, then the linear stage is added to provide better load step regulation; simply to reject voltage dipping and overshooting in the range of a few kilohertz. You don't even expect it to reject switching noise (overtones) or even the fundamental switching frequency.

Still, designing a quiet switcher is very important as it is nearly impossible to filter out the crap that a mediocre SMSP gives out. It goes through the parasitics of any filter.

[BravoV]

Just a suggestion, why not take a look at this circuit or even buy the kit  , quite pricey though.

DC2132A - 24V 3A Constant Voltage, Constant Current Bench Supply , details page here -> HERE or it's performance (PDF) -> HERE

Imo, just copy and build it rather than have to go through the time, effort and costly development stages. Unless the OP's intention is to learn through the process.

It was briefly discussed and reviewed by one of the forum member on it's performance like transient response, output noise etc -> Here

EDIT : According one of the Linear Tech staff (I presumed), the BOM cost for somebody doing this might be sub $50.

[Kleinstein]

The LTC circuit is not using a filter between SMPS part and linear part - this is a rather effective way against the difficult to handle high frequency "Noise".