[time-nuts] Power supply for time source concerns

[jimlux]

On 12/23/19 5:01 AM, Achim Gratz wrote:
> Am 22.12.2019 um 20:37 schrieb Taka Kamiya via time-nuts:
>  > Most commercially manufactured time and frequency sources use
>  > switching power supply.
> [...]
> 
> The suggestion to "just use a linear power supply", especially if it was 
> designed some decades ago is probably not getting the desired results 
> these days.  There is a lot more conducted noise on the power lines than 
> there used to be, starting from several kHz right into RF territory. 

An excellent point - The plethora of "offline SMPS" such as phone 
chargers, and various and sundry USB power supplies, means that even if 
they have reasonable conducted emissions, there's a lot of them all 
contributing.


> Linear regulators generally have bad PSRR at higher frequencies (often 
> starting to degrade in the low kHz region already) and unless you have a 
> clean input supply you pass any HF noise almost directly to the load. 
> SPS are not necessarily worse than linear supplies when it comes to 
> noise, but it generally shows up in different places frequency-wise (the 
> detailed characteristics depend a lot on the exact topology chosen, so 
> that's a whole 'nother dimension of things to consider when chosing). It 
> is also harder to contain the switching noise as you have several 
> high-current loops typically whose area must be kept as small as 
> possible to not radiate noise.  Conducted noise can be more easily 
> filtered, but it gets out both through the input and the output side. 
> The input side is often neglected a bit, which can come back to bite you 
> when you have multiple supply rails in the system.  Most switching 
> topologies (or at least the ones that can be "clean" enough for the 
> purposes under discussion) will produce a triangular ripple at the 
> output at the switching frequency.  You can make it smaller and smaller 
> at the expense of load regulation, but it usually is easier to just deal 
> with whichever number of millivolts that you're left with by putting an 
> LDO post-regulator directly at the load (you put it at the load so you 
> have better regulation and you can also use the supply line between the 
> SPS and the LDO as part of a Pi filter).  The LDO needs to have good 
> PSRR at the switching frequency and maybe the first or second harmonic, 
> the rest of the spectrum should already have been dealt with by filtering.

It is tough to build a "small" input filter that has good rejection at 
low frequencies (<100kHz?)

What we wound up doing for a low noise design was cascade the following:

Unregulated 9-15V bus
input filter
Buck converter to 8V   (boost converters seem to have more noise)
Buck converter to "usage voltage" + 0.5-1V (i.e. if it's destined for 
3.3V, buck down to around 4-4.3V)
Pi filter with 60 dB "design" attenuation (probably less on the board)
Board to board connector
Pi filter with 60 dB "design" attenuation
LT3042 LDO to load.

We did some experimentation to measure the voltage drop across the 
filters, and then adjusted the output of the second buck converter so 
the LDO would have enough headroom over temperature.


The layout was 4 layer boards, nothing exotic, other than the usual 
separation of pre & post regulation power & ground planes.



> 
> 
> Late Jim Williams' (RIP) application notes are always a good read:
> 
> https://www.analog.com/media/en/technical-documentation/application-notes/an70.pdf 
> 
> 
> https://www.analog.com/media/en/technical-documentation/application-notes/an101f.pdf 
> 
> 
> https://www.analog.com/media/en/technical-documentation/application-notes/AN118fb.pdf 
> 
>