RFC: Switching mode power supply schematics

[nitrousnrg]

7.5KHz is a quite low frequency. I was expecting to see a much higher freq noise.

Anyway, whats the input voltage in your simulation?

[KW1252]

It was probably some sort of oscillation, or software error. Input is 13.5 volts with plenty of noise. However, if the next graph is a valid result and not a software malfunction or Bad User Error, it's the jackpot.

[nitrousnrg]

wow, very nice.

Time to ask octopart?

[KW1252]

Here's the circuit:


And the exported BOM. There's a lot of no good BS like having a 2N2222 transistor in the LDO part, the only components that did matter were the RT pin resistor, VC RC circuit, and obviously everything connected to the SW pin. The 1k resistor and 1u cap is a simulated load with decoupling cap. Schottky isn't the same as in the reference circuit, I tried to choose a "close enough" diode for it. All the components should be readily available.



--- Bill of Materials ---

Ref. Mfg. Part No. Description
C1 Nichicon UPG1H100MPH capacitor, 10µF, 50V
C2 -- -- capacitor, 1nF
C3 KEMET T510E687K006AS4115 capacitor, 680µF, 6.3V
C4 AVX TAJB226M006 capacitor, 22µF, 6.3V
C5 -- -- capacitor, 1nF
C6 TDK C3225X5ROJ226M capacitor, 22µF, 6.3V
C7 -- -- capacitor, 220nF, 50V
C8 -- -- capacitor, 1nF
C9 KEMET T530Y337M006ATE006 capacitor, 330µF, 6.3V
C11 AVX TAJA105K016 capacitor, 1µF, 16V
C12 Panasonic EEFUD0D221R capacitor, 220µF, 2V
C13 TDK C1608X5RIA105K capacitor, 1µF, 10V
D2 Fairchild SS24 diode
L1 -- -- inductor, 64nH, 3A pk
L2 Wurth Elektronik 7443556560 WE-HCB 1890 inductor, 5.6µH, 33A pk
L3 Taiyo Yuden NR3015T-1R0N inductor, 1µH, 2.1A pk
Q1 Philips 2N2222 bipolar transistor
R1 -- -- resistor, 1K
R2 -- -- resistor, 57.6K, 1%, 0.1W
R3 -- -- resistor, 10.2K, 1%, 0.1W
R4 -- -- resistor, 30.1K, 1%, 0.1W
R5 -- -- resistor, 57.6K, 1%, 0.1W
R6 -- -- resistor, 10.2K, 1%, 0.1W
R7 -- -- resistor, 10K, 1%, 0.1W
R8 -- -- resistor, 10K
U1 Linear Technology LT3694 integrated circuit

[KW1252]

Strange the BOM didn't mention actual data on L1, it's not just "inductor", it's Wurth Elektronik 742 792 112 Ferrite bead,64nH, 3Apk.

The extremely nice curve WAS partly a bad user error though, at 1k load the circuit is practically at idle. With 3 ohm load (1,7A) the ripple/jitter goes to about 150µV. Still quite manageable but not as nice as it seemed. Also Vout drops to 4.9 volts.

[KW1252]

Here's the simulation with load applied:



that's about 160µV from the lowest peak to the max. Should be still plenty good for most IC's though. There's obvious harmonic pattern of 100Hz , plus less obvious at 1500Hz. I wonder if a squiggly trace before chip power pin and decoupling diode would provide enough inductance to filter any of that out.

Another thought I had; how about if the switching mode power supply is the always on system, and analog would be the one to turn off, plus the power hungry component (e.g. LM1949's) would be shut down by a common FET between VSS and ground. It would probably require protection diodes though.

[KW1252]

I made a reference circuit with the current power supply schematics as a reference with a separate LDO regulator (LT1086-5 as, suprisingly, a Spice supplied by Linear doesn't include IC's from other manufacturers), LDO built with the driver built in the LT3694 with the same configuration, and the filtered switched output. Here's the results:



Question: which is which?

Answer: Green graph at the top is the built-in LDO of 3694, red at the middle is the separate 1086 LDO, and the one in the bottom is the filtered output from the switched power supply. To be fair though, the top red is exact replication of the Vin noise, the separate LDO does respond to the noise but can't filter it out. Going into extreme detail, the switched output does have some noise:



However, amplitude being 30µv I hardly consider it a problem. The test was made at very light load, the load resistor for every output was 1k.

Going into high load on the switched mode power supply, the noise begins to show up on the switched power out:



The LDO's are still at 1k, as they serve best in the sensor feed which never requires much current. The closeup reveals an interesting thing:



The noise is the same shape as the input noise. The benefit here is when you filter the input for LDO's, you also clean up the switched power output.

While simulations won't tell the entire story, they do give you at least the ballpark estimates. With these results, I feel very confident about going forth with the switching mode power supply. I even daresay it might improve accuracy, as heat is also a source of analog errors.

[jharvey]

I think the acceptable wonder from 5V/3.3V should be around 1mV or less under any condition. Whether it be ripple, temperature, load, ect, we should strive for under 1mV. Does that sound like a good general design spec? To me it looks like the above meets that spec.

Also lets not forget that ADC have issues if you don't use radiometric and non-radiometric configurations quite right. So concern of supply induced power supply ripple noise could be a mute point if the ADC's are configured correctly.

[KW1252]

The ripple is well below 1mV, and with voltage input filtering it gets even better.

Here's the overview:

Purple is the simulated noisy charge voltage from the alternator. Dark blue is after filtering; the values are WAY off, but this is the first attempt to iterate stuff


Voltage at the outputs, the light blue is the output from 3694 LDO driver, green is the 1086 LDO, and the red is again the 3694 switched. The 3694 LDO clearly is the worst of the bunch, as it's meant to be a step-down regulator from the switched output. With heavy duty filtering it might be useful, but then again a separate LDO would behave better and possible require less components on the board. Then the 3694 should be replaced with a switching regulator without the LDO driver outputs, as they are waste of PCB estate and power.

Here's finally the circuit used:

[nitrousnrg]

I'm setting up a digikey order in a couple of days to try my SMPS (and other freeems-unrelated things), and I'd like to have yours too to compare.

I had your values added to my list of components, but I bet that if you change the IC for a non-LDO one all the passives are going to change too. So, if you think it can be of help, I can test the posted circuit, or a new one as long as its posted before saturday :-)

Checking the order, it looks expensive, but we'll see.