Roast my design... Low noise DC supply

[tchiwam]

https://github.com/tchiwam/LNA



1st time PCB design since a long long time, comments are welcome... Still on the fence to add mounting holes and a resetable fuse.

[Simon]

so is that actually about 80% wasted space there or are some parts not showing?

[Kleinstein]

The linear regulators are supposed to get hot when in use, especially for higher current, as suggested by the high current version.
So the regulators should be separated a little more and maybe have space an holes for a heat sink.

There is relatively little filtering on the input side.With a classical 50/60 Hz transformer there should be space for electrolytic caps. With a SMPS it would be a good idea to have more LC filtering. Supply from a battery may want a fuse.

This is not an RF design, so no need for all those vias.

The regulators may work with just ceramic capacitors at the output, but often stability gets better if there is also some capacitance with loss / ESR: either an electrolytic cap or a cap with a small series resistor.

[tchiwam]

Thanks for the feed back.

    This is my take on the page 27 of the LT3045/3042

     The spacing can be improved, and the heat sink will be provided by the external case too. RF style box to match the LNA, both will be kept at -40C.

Board dimension  50 x 25mm = 1250mm²  is to keep with the LNA form factor and make is easy to support mechanically. The datasheet asks for 50x50mm = 2500mm² per regulator. I hope the machined top and bottom case will help a little.  My max load will be about 100mA at 5VDC. The chiller at -40C may also help.

I picked X7R capacitors and added the 22µF big package from the design of someone else. I may or may not use that big, the performance of the single chip was close to the datasheet specs, will have to see for this version.

Input filtering: as per design guide, they ask not to put input caps too close to the chips, but I can add some near the input.

Output filtering: same story, ceramic cap, -40C use, as per design recommendations

Todo:
   Regulator spacing
   Adding an input fuse
   rework vias near regulator
Maybe:
    50x50mm board version, need to rethink input and output traces

Simon : https://www.analog.com/media/en/technical-documentation/data-sheets/LT3045.pdf page 22 is part of the reason for all that empty space.

[tchiwam]

Correction: Board size is now 50x100mm, so 5000mm² so half of the recommended 2500mm² per chip...

[tonyh88]

+1 for all the points in Kleinstein's answer

You have the room anyway so you might as well use it and spread the regulator on the board. Unless there are mechanical constraints

[tchiwam]

The 20mohm ballast resistor traces are kind of a limit, so are the input power planes.

I've added a bit more vertical space between them.

[ElectronRob]

Can you share some info regarding the design spec - what are you doing, what does this board need to achieve?

Do you need any common mode performance?
What is supplying the input? It better be a battery.
What is the ambient temperature, you mention -40, are you going to cool this?

You say 1.8-20V input, and 0-15V output @ 2A. So with 20V in and say 1V out your asking a lot even with the copper area you have.

If you are using this to power a LNA what are the spec for that? Bandwidth etc, what are you amplifying (differential/single ended)..... Apologies if you have another thread somewhere but I think you would get more help with some more info plus I (and I'm sure others) find looking at a schematic useful regardless of how simple the layout is.

What are you hoping to achieve with the vias btw? Just better thermal performance?

[tchiwam]

ElectronRob: Thanks for the hints,

I added the schematic in pdf here in case someone doesn't want to install Kicad : https://github.com/tchiwam/LNA/blob/main/Kicad/LT3045-4.pdf
All of it on github : https://github.com/tchiwam/LNA/

This is my 1st complete board design with Kicad, so far, yup some akward, but I like it more and more.

Here a new shot at this: seems the 1st post is updated since this is an external link...

The goal is to get the lowest Vrms noise, I was planing to handle the Common mode outside of the design, but added a small CMC (L1).

Mainly power for ADC and LNA.

The power output will rarely go above 100mA @ 5VDC.

Input power will be well behaved, isolated power supplies. I'd love to see if a battery would help here.

I plan on adjusting the input power accordingly, never above 8VDC above output voltage. (Hmm, will add a not on the silk screen)

The power supply will be kept at -40C, same as the LNA in this case. Single end, right on the horn, as wide band as we can get from 0.6 to 6GHz. But probably in steps with a noise source in the path.

The signals I am interested in are around -140 to -155dBm, and some big noisy motors will be present around the antenna, common mode hello.

I simply followed all the bits Analog Device suggested in their datasheet and app notes. They do not have a reference design for the 4x LT3045/LT3042   so this is my take on it on all the madness.

[tchiwam]

Here is the bare :

[tchiwam]

https://www.analog.com/en/products/lt3045-1.html#product-overview  and yeah ... Found a design using the LT3045-1 ... the 12 pins variant 

[asmi]

These LT3045's are super freaking expensive. TI has recently released a pair of switchers (2A and 3A) specifically for low noise applications: https://www.ti.com/product/TPS62913 They promise <10uVrms, which, if true, is amazing. As I understand, they've managed to stabilize feedback loop even with the ferrite bead in the loop, so they "sample" voltage after it, and that compensates for the losses on bead. These parts are in preview/ES status as I understand, but you can order up to 50 devices through their website.

[ElectronRob]

Haha, I read that and though pffft. can't be that expensive... I was wrong 

[tchiwam]

about 14$ each regulators ... Might as well try to make the full design : 800nV ripple is interesting for my application.

[ElectronRob]

Ok to the design.

Thanks for the extra info. I will look at the schematic in a moment.

First some dull comments.

1) If this is going to be used by someone else (other than you) make all labels explicit, that is to say. Don't say output from 0-15Vdc with a current together with a max 20V input. They will assume its fine for that and proceed to make all the magic smoke come out. Rather give a power rating, or decrease the maximum output current silk so that with the max input V your device temp does not exceed its maximum. Note also that the maximum output current (or power) can only be achieved at the -40deg design spec - cover your arse, then use it however you like when your driving.

If you want to be really fancy you can add various lumps of protection to make sure no one can exceed ratings - various methods should you need them.

2) Although this is likely in a box, its still nice to add some low power LEDS on the input and output, especially if you later add a re-settable fuse or efuse type device. I've used the LM5069 a lot recently, now chuck them into most designs by default and they happen to be rated down to -40 (although haven't read the data sheet to see if there are any low temp caveats).

I'm just off for some dinner (she has started to shout) so will look at your schematic when back, but your intended frequency is useful info. I will leave by suggesting the addition of a low pass filter, you will find you can likely reduce the output noise in the frequencies you are interested in. I would suggest that you run your filters (future and present) through TINA or LT spice to check they don't have any higher frequency horrors as you often will with RL. LTspice also has a nice noise tool which I've found to be of help with low noise projects in the past.

Expect you will also be adding a Vin RC for each LNA stage locally too?

[tchiwam]

At 1st I wasn't sure I'd add the current limit, since it was rather simple and out of the way (layer 2 Vertical bus) I added the few resistors.

Now I looked at the Analog design for the 3045-1 (12 pins instead of 10, might resping a PCB for that one too)

They T the Vin, rather simple. I am trying to keep the distance from the Input to the regulator as mentioned in the design suggestions. So I went for a fork with small ground bridge near the regulator vias.

Comment #1  :  I added the 260mV < (Vin - Vout) < 8.5V thinking of your comment. This is solely for experienced users, mainly me and very few others, the silk screens are just to allow others to spin their heads around it and not smoke the expensive 3x3mm black thingies...  I will implement the reverse protection and other goodies that are included in those regulators, only if they don't lower the quality of the power.
 
Comment #2 :  Yeah, the conflict is real Diodes make noise and heat   I may just add one on the Input and use the PG pin 4 for a led.

LP filters, I use Qucs for filters, LTspice was one of my old favourites   TINA I don't yet know of.

I have the QPL9065 in mind for the 1st LNA...

[asmi]

about 14$ each regulators ... Might as well try to make the full design : 800nV ripple is interesting for my application.

800 nV ripple is very likely a massive overkill for your application. Also these parts like to run super hot.
I've recently completed an RF project, which started with a ton of LT3045s , but the final revision ended up with TPS7A92's + bucks, which are like 5 times cheaper, and are still good enough for application, and the board runs cool, while first prototype was +55°C (it's rather large 4 layer board with solid power and ground planes).
Over-engineering is rarely a good approach for anything other than early prototype.

[tchiwam]

0.4µV is about -115dBm if it would be right in 50ohm, I will shove it in a DC block to my spectrum analyzer and see where it Goes.

This will be a fun experiment, or at least a place to start. Our next step is LNA on PCB @ -40C, then wire bonding and colder. I know the can of freeze it is already doing wonders on the LNA

I do plan on keeping the supply near the output voltage in order to manage heat dissipation. But for us it might be great to shove the hot side on the opposite side of the LNA chiller and keep condensation away.

Good to know the 3045 are running hot, they are linear, so I expect 50% loss in heat (at least)

LNA ~500mW  , LDO ~1W , chiller 80W, should be good.

EDIT: I expect a T-rise of about 30 to 40C

[ElectronRob]

At 1st I wasn't sure I'd add the current limit, since it was rather simple and out of the way (layer 2 Vertical bus) I added the few resistors.

Now I looked at the Analog design for the 3045-1 (12 pins instead of 10, might resping a PCB for that one too)

They T the Vin, rather simple. I am trying to keep the distance from the Input to the regulator as mentioned in the design suggestions. So I went for a fork with small ground bridge near the regulator vias.

Comment #1  :  I added the 260mV < (Vin - Vout) < 8.5V thinking of your comment. This is solely for experienced users, mainly me and very few others, the silk screens are just to allow others to spin their heads around it and not smoke the expensive 3x3mm black thingies...  I will implement the reverse protection and other goodies that are included in those regulators, only if they don't lower the quality of the power.
 
Comment #2 :  Yeah, the conflict is real Diodes make noise and heat   I may just add one on the Input and use the PG pin 4 for a led.

LP filters, I use Qucs for filters, LTspice was one of my old favourites   TINA I don't yet know of.

I have the QPL9065 in mind for the 1st LNA...

QPL9065 looks nice - their EVAL board also looks well done and something which I would be copying. They make provision for input filtering with the 0R also.

I would be placing the LM5069 (if you end up using one at all, if its just you hooking this up then perhaps no need at all) right on the input before any filtering, with a 1u on the timer pin to mostly disable it but am often charging a large down stream capacitance which you won't - they have a nice excel spreadsheet to work all the values out for you I usually use BSZ520N15NS3 G as the series fet, tough, small, yet to make one into toast. At least gives you some level of protection (stupid mistakes happen to the best of us) for later expensive and no doubt fragile LNA's.

Yes good point regarding LED heat (however the rest of the cct will be making enough), in this case I would break connection to the LED so you can short a couple of pads out with tweezers while testing, a little LED goes a long way when you are stressed and trying to figure out why something isn't working, your meters hooked up to something else blah you know the score. Also helpful when trying to talk someone else through diagnosing a problem down a phone (thats never fun), good sanity check - is the LED lit?

Ultimately given your bandwidth, with some simple filtering I suspect you are going quite a way overboard, but there is merit and fun in that. As a silly example, with a series 1R and 10u cap on the output you are down over 50dB by 10Meg. Exchange the 1R for a well selected ferrite bead and you'll be closer to 100dB down, perhaps more. Suddenly your not looking so bad with an opamp and a large BJT - horrid and cheap.

I don't know what your total set is like but you may want to consider some common mode filtering on the output. I would be worried that even through you look to be floating(?) you will end up with some CM to diff conversion. I have a long term project where a very well balanced pick up coil with high CMRR is upset by a large transmit from another coil, the 'noise' should all be common mode and I am fully differential however I still get cross talk due to some tiny imbalance hence the CM begins to matter, usually more at high frequencies.

I have to say I don't ever design at frequencies above 100Mhz (ignoring clocks etc) so someone with real RF experience would be valuable here.

Sim your inductor/capacitor filters, you may well find you require a low value series R to damp. Also note that large package components do not like thermal cycling, so rapid cool down, or more likely taking them out of the -40 into ambient could ruin your day.

No harm making any thermal vias larger, Saturn PCB has a nice via calc showing thermal resistance of vias.

Not sure I have more practical internet advice to offer. Best of luck with what looks like a fun project. Will be watching with interest hoping that an RF wizard flys through with some old school advice.

[ElectronRob]

As a final closing thought, do you need this power supply in the cooler? Seem horribly inefficient.

[tchiwam]

As a final closing thought, do you need this power supply in the cooler? Seem horribly inefficient.

It will help to hold everything in one block and keep the DC lines very short between the LNA and Regulator.

Separating them will be harder to isolate too...

I will have 80W of cooling, so might as well cool another 500mW. 

[asmi]

0.4µV is about -115dBm if it would be right in 50ohm, I will shove it in a DC block to my spectrum analyzer and see where it Goes.

This will be a fun experiment, or at least a place to start. Our next step is LNA on PCB @ -40C, then wire bonding and colder. I know the can of freeze it is already doing wonders on the LNA

I do plan on keeping the supply near the output voltage in order to manage heat dissipation. But for us it might be great to shove the hot side on the opposite side of the LNA chiller and keep condensation away.

Good to know the 3045 are running hot, they are linear, so I expect 50% loss in heat (at least)

LNA ~500mW  , LDO ~1W , chiller 80W, should be good.

EDIT: I expect a T-rise of about 30 to 40C

I guess I didn't phrase it well enough. These extremely low noise LDOs need to be physically very close to the IC they power (to avoid picking up noise) - so they need to be on the same PCB as whatever they provide power to. That's why I said that in your case you will not see any benefit whatsoever from them, and it's going to be nothing more than extremely stupid waster of money. As a matter of fact, I'm willing to bet that if you build two variants of the same board - one with LT3045, and another - with aforementioned TPS7A92 (or any low-ish noise LDO), you will not be able to measure any difference. Hence - "stupid waste of money".
Take a look at any PCB with sensitive analog components - you will see that they tend to have local power supplies which are ALWAYS right next to power sinks, and NEVER on external boards.

[tchiwam]

Asmi, we will eventually merge the LNA. That's the plan. If we settle on a LNA that runs on 1.8 to 5VDC, I may just make a new board with that TPS7A92.  Able to provide a 12V out is a nice thing too.

The proof of the puding is in the eating

[tchiwam]

QPL9065 posted in the RF Microwave forum...