Hi, I've a six channel DAC project that's planned to drive a pair of three channel amplifiers. The supplies for the DAC and its supporting circuitry are 2.5V 25mA, 3.3V 75mA, 5V 175mA, and -5V 100mA. The supply for each amp is +-7.5V 2A, though the control loops run on +-5.7V 15mA regulated. The 2.5V, 3.3V, and 75mA of the +5V supply are inflexible as they're dictated by the design voltage of the digital or DAC circuitry consuming the supply. The 7.5V, 5.7V, and 100mA of the +-5V rails are analog and are more free to vary, though decreasing them doesn't work so well and increasing them significantly runs into thermal limitations as the analog circuits are mostly medium pitch surface mount.
For a variety of layout, modularity, mechanical, and aesthetic reasons I would like to implement the DAC and amps on separate boards but have the option to share a transformer and bulk capacitance between the DAC board and one of the power amps. This is easily done with a linear supply. The current demand's a little high for 317+337 regulation from a 9V trafo but a pair of LM350s or LM338s in TO-220 with some board mount heatsinks would do the job of flattening the mains variations into +-7.5V just fine, enabling compact point of load regulation for the other supplies without needing to fuss much with thermals. It's also feasible to use a 6V transformer to deliver +-7.5V unregulated; while this makes delivering the negative rails from positive regulators tricky the two negative LDOs needed to pull this off while maintaining stability with MLCCs do exist. However, neither of these options is especially attractive to me. A linear supply from a 9V trafo ends up less than 50% efficient in delivering power to the loads and the 6V trafo is kind of sketchy as limited negative regulator availability leads to some design corners with maybe 50mV of margin before the regulators go into dropout.
So I find myself looking at switchmode solutions, nominally with the intent of of providing +-5 and +-7.5V regulated rails and deriving +2.5 and +3.3 from +5 at point of load. The design already calls for a CLC pi filtering to separate loads from each other on a rail. So 100+dB of switch transient rejection is mostly built in already and I've no real concerns about loss of DAC, op amp, or power amp performance due to switching on rails. I am not deeply versed in SMPS solutions or parts. But it appears there's no such thing as a +-7.5V dual supply module and approximating one by pairing a +-5V module with a +-9V module looks to start at around USD 120 if one wants decent bricks with a high enough switching frequency for good pi filter rejection. However, that would mean spending about as much on the supply modules as all of the other parts in the system combined. That's a bit steep for a not particularly ideal solution. So I've been looking into building my own SMPS. It seems like buck conversion is the most cost effective and simplest option here but I'm having a hard time finding modules or regulators which maintain good phase margin with output capacitances of several hundred microfarads---it's desirable to use polymer electrolytics here due to their low cost, low ESR, low ESL, and strong tendency to be available in sizes large enough to enable low cost, low ESR, and reasonably comapct inductors to be used in CLC filtering.
The best candidate parts I'm finding are TI's SimpleSwitcher constant on time LMR regulators and LMZ modules, which I've been simulating in Webench. The DAC supplies reduce to +5V 250mA and -5V 100mA class A loads and seem not especially difficult to accomodate with an LMR12010 or LMZ12001 due to the minimal motion of the load pole. However, the power amp swings from 45mA quiescent to 2A max and can go as low as 24mA if something bollixes up and the output devices go into thermal shutdown. This seems to be more load pole motion than the LMR parts can accomodate and looks decidedly marginal with the LMZ14202H, which is the module which best satisifes the requirements. As a caveat, I'm not sure how much to trust Webench simulations for discontinuous conduction mode; I'm getting strange results like phase margins going to -90 degrees at low frequency and then swinging back to +80 where the gain crosses 0dB.
My questions to the forum are therefore pretty simple. Am I looking at the right topologies and parts or is there a better way to go about this? If so, are the odds of success good enough to be worth trying an implementation? If not, what should I be looking at instead?
For a variety of layout, modularity, mechanical, and aesthetic reasons I would like to implement the DAC and amps on separate boards but have the option to share a transformer and bulk capacitance between the DAC board and one of the power amps. This is easily done with a linear supply. The current demand's a little high for 317+337 regulation from a 9V trafo but a pair of LM350s or LM338s in TO-220 with some board mount heatsinks would do the job of flattening the mains variations into +-7.5V just fine, enabling compact point of load regulation for the other supplies without needing to fuss much with thermals. It's also feasible to use a 6V transformer to deliver +-7.5V unregulated; while this makes delivering the negative rails from positive regulators tricky the two negative LDOs needed to pull this off while maintaining stability with MLCCs do exist. However, neither of these options is especially attractive to me. A linear supply from a 9V trafo ends up less than 50% efficient in delivering power to the loads and the 6V trafo is kind of sketchy as limited negative regulator availability leads to some design corners with maybe 50mV of margin before the regulators go into dropout.
So I find myself looking at switchmode solutions, nominally with the intent of of providing +-5 and +-7.5V regulated rails and deriving +2.5 and +3.3 from +5 at point of load. The design already calls for a CLC pi filtering to separate loads from each other on a rail. So 100+dB of switch transient rejection is mostly built in already and I've no real concerns about loss of DAC, op amp, or power amp performance due to switching on rails. I am not deeply versed in SMPS solutions or parts. But it appears there's no such thing as a +-7.5V dual supply module and approximating one by pairing a +-5V module with a +-9V module looks to start at around USD 120 if one wants decent bricks with a high enough switching frequency for good pi filter rejection. However, that would mean spending about as much on the supply modules as all of the other parts in the system combined. That's a bit steep for a not particularly ideal solution. So I've been looking into building my own SMPS. It seems like buck conversion is the most cost effective and simplest option here but I'm having a hard time finding modules or regulators which maintain good phase margin with output capacitances of several hundred microfarads---it's desirable to use polymer electrolytics here due to their low cost, low ESR, low ESL, and strong tendency to be available in sizes large enough to enable low cost, low ESR, and reasonably comapct inductors to be used in CLC filtering.
The best candidate parts I'm finding are TI's SimpleSwitcher constant on time LMR regulators and LMZ modules, which I've been simulating in Webench. The DAC supplies reduce to +5V 250mA and -5V 100mA class A loads and seem not especially difficult to accomodate with an LMR12010 or LMZ12001 due to the minimal motion of the load pole. However, the power amp swings from 45mA quiescent to 2A max and can go as low as 24mA if something bollixes up and the output devices go into thermal shutdown. This seems to be more load pole motion than the LMR parts can accomodate and looks decidedly marginal with the LMZ14202H, which is the module which best satisifes the requirements. As a caveat, I'm not sure how much to trust Webench simulations for discontinuous conduction mode; I'm getting strange results like phase margins going to -90 degrees at low frequency and then swinging back to +80 where the gain crosses 0dB.
My questions to the forum are therefore pretty simple. Am I looking at the right topologies and parts or is there a better way to go about this? If so, are the odds of success good enough to be worth trying an implementation? If not, what should I be looking at instead?
Seems the preferred part would be the LMR24220, though the LM26003 would work too if followed by a linear regulator to suppress audio band ripple in DCM. The LMR24220 wants post regulation as well. The switching frequency at my minimum load is 62.5kHz, which is not so easy to reject with an LC filter cornering above audio frequency, and Webench is giving nearly 500mV of output swing on a 200kHz load transition (the slowest Webench allows) which is actually worse than the performance on faster edges and about three times the ripple of an unregulated supply. So the 6V transformer seems the best option despite the marginal Vee corner (for the less than the price of a LMR24220 one can get nice low power LDOs to regulate the amp's control loop; saves board space too).
I decided to revisit this to see what had changed in the past year and came across AN-1141, wherein Analog obtains higher SNR and SFDR from LC postfiltered buck converters than with LDOs. (AN-1141 was actually published in 2012 but I missed it last year.)
Some notable buck parts delivered this year are the CS51414, NCV890200, TPS54340, and TPS54540, all of which support sufficiently high minimum switching frequencies and sufficiently low ripple an LC filter with an audio passband can reduce switching noise to a few microvolts. That's typical of noise floors in audio circuits so is probably good enough. They also support inverting topologies and LM3x7/LM7xxx rail voltages and hence are audio dual supply friendly.
The TPS54x40 hold up well on a look in Webench for positive rails (inverting is still not supported)---80C Trise at my worst case corner, 43kHz regulation bandwidth with 50 degree phase margin, well behaved in DCM, with the sims claiming 4mV of output ripple. Inverting regulation bandwidth's usually significantly lower for buck converters but this is good enough Vee performance could be acceptable. I've not done the loop maths as yet but will likely do so.
Some notable buck parts delivered this year are the CS51414, NCV890200, TPS54340, and TPS54540, all of which support sufficiently high minimum switching frequencies and sufficiently low ripple an LC filter with an audio passband can reduce switching noise to a few microvolts. That's typical of noise floors in audio circuits so is probably good enough. They also support inverting topologies and LM3x7/LM7xxx rail voltages and hence are audio dual supply friendly.
The TPS54x40 hold up well on a look in Webench for positive rails (inverting is still not supported)---80C Trise at my worst case corner, 43kHz regulation bandwidth with 50 degree phase margin, well behaved in DCM, with the sims claiming 4mV of output ripple. Inverting regulation bandwidth's usually significantly lower for buck converters but this is good enough Vee performance could be acceptable. I've not done the loop maths as yet but will likely do so.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.