Discerning FL2K devices with LDOs versus switching regulators
Greetings, In Steve Markgraf's slide presentation (http://people.osmocom.org/steve-m/fl2k_slides/osmo-fl2k.html), do slides 16 and 17 imply that some FL2K devices have LDO regulators while other using switching regulators? Obviously, the FL2K devices that have LDO regulators are preferred, due to fewer spurious RF emissions. How can one determine which FL2K devices have LDOs? Can an FL2K device be reworked to use LDO regulators? Thanks, Mac
Hi Mac, so first of all: any spur is only a problem if it ends up in your signal. Since we're clearly talking about devices that you can't use for operation with an antenna withou very much filtering: check whether you actually get a problem first. To be completely honest, the whole LDO vs. SMPS discussion often bares technical background, as you'll find SMPS in high-end radio receiver devices just as well. It's all about /designing/ your thing to be low noise, not about the "use an LDO instead of a switcher". Now, Steve has offered nice figures about the spurs there, so these might actually be linked to the switcher. However, the method seems to be to first measure, then link cause to it; not the other way around. I'd argue that the device with the many spurs that, actually do look like one rectangular wave modulated another rectangular wave, was simply badly designed, probably with underdimensioned means of eliminating cross-talk between the two switchers (no idea how they relate). What confuses me is that these spurs roughly fall into a 3 MHz grid – and that's usually a bit on the high end for switching frequencies. Another device with a switcher *might* be nicely filtered and work perfectly well. I agree, adding a switching regulater definitely adds a source of noise, but please don't assume that cheaply designed LDO systems are superior in signal quality¹; there's modern switch mode supplies that actually use spread-spectrum methods to spread out the energy they leak onto many frequencies², and others that you can synchronize e.g. to sampling clocks so that noise at least aligns and can be filtered out more easily. The point I'm trying to make is that if these spurs are a problem to you (and I can heartily to figure on slide 17 worrying you), then you'll want to have spur measurements at different sampling rates at exactly your USB bus – in the end, the noise of a SMPS very much depends on how hard it is at work, and a stable input supply and high output current might be nicer than a dropping input and a current draw so small that forces the SMPS into discontinuous current mode. Regarding spotting: ------------------- Switch mode supplies generally can be found by looking for (large-ish) inductors close to (large-ish) diodes, typically close to either a converter IC or in higher-current applications close to a (large-ish) discrete transistor. Do an image search for "SMD power inductor", and you'll see how these tend to look like. Regarding remedies: ------------------- Filters, filters, filters³! You need to select the right Nyquist zone, anyway. So, pick a sampling rate range that works out for that; shifting your signal in digital domain so that it ends up where you want it after being shifted by the sample clock N times allows you to have some leeway there. Then, use whatever remaining degrees of freedom you have to pick a rate that is at a supply spur – and filter that out. Whether this is an option at all of course depends on the RF bandwidth you need. Replacing the power supply on-board: I'm willing to say "it's possible", but I'd also say "at a time investment higher than simply buying a handful of candidates and simply sticking with one that works". Supplies typically have to be electrically well-coupled to the ground and supply lines, so if you externalize these, you'd replace the original output stage of the on-board SMPS with larger capacitors, but these typically have worse RF interference suppression properties, so you'd add smaller capacitors, but now you have a system with capacitors of different sizes and internal resistances and inherently some inductive characteristics of whatever connects the external supply to these connectors – you can certainly simply build that, and it's not that unlikely it'd work, especially if you overdimension everything a bit, but I wouldn't know how to predictably make a "first trial works" device. Note that switch mode ICs for these voltages and currents aren't necessarily solder-friendly⁴ . Rule of thumb: The smaller the package, the higher the switching frequency⁵ – and as noted above, 3 MHz would be at the higher end of the spectrum of switching frequencies⁶, but that's likely because higher switching frequency also makes the necessary inductance smaller, and hence, the inductor cheaper. What I would do =============== Compare a handful of dongles. Because: a) They're cheap, and time is sparse, b) can't be that bad to have spare ones lying around, for operation away from the spurs, or to give to friends who want to try that, or to honestly resell as tested to work with osmo-fl2k but replaced with a lower-noise one, c) to verify hypotheses on how to fix things, without risking to fry one of the "good ones", and c) if you can figure out how to improving the best one, maybe it becomes easy to improve the others, too. Maybe it's easier to observe an improvement in the ones that are bad. Go and measure. That means that I'd both add appropriate output filters for both the Nyquist zone I want, and measure after that (e.g. using an RTL dongle, whose spurs I at least know), as well as trying to figure out where exactly the spurs come from – are they really on the signal lines, or are they radiated into my measurement by the shield conductor of the VGA port? When I probe around with an oscilloscope, on which lines do I see exactly these frequencies I observed? Then, improve and adapt. If things are actually radiated by the board, proper shielding might be the simplest method to improve the situation. Else, go for easy things like soldering another (better, as in lower ESR, higher capacity?) capacitor onto the decoupling capacitors or output smoothing caps on-board⁷ first. Best regards, Marcus ======================================================================= ¹ You can underdampen these LDOs, just as well, or underdimension them: linear supplies tend to be cheaper than SMPS for small loads, so the fact that some manufacturers use SMPSes might point out that you'd need a relatively beefy and fast LDO and thus expensive LDO to reliably supply the current needed, and there's plenty that you can mess up when you're designing an LDO system at the edge of cost efficiency ² Though that doesn't sound too desirable here ³ Imagine Ballmer going "developers!" on you here. ⁴ In highly integrated electronics, ICs with 6 pads in a package of total size ~ 1 mm × 1.3 mm would be typical if you just need a small step-down from 1.8 V to 1.2 V efficiently. ⁵ Because the higher the frequency, the less charge transfered per cycle, the lower the switched current, the smaller the switching transistor. ⁶ Please don't really infer that this means you get a chip scale package – these VGA dongles were built with cost, not size, as primary target, as you can see from the sparsely populated simple PCBs; you don't use a high-end phone-building assembly line to build 5 € VGA dongles, so you don't use <0.05 mm tolerance in placement parts. ⁷ Maybe don't take it this far: https://twitter.com/LaF0rge/status/892872883164336128 On Wed, 2018-05-09 at 22:44 -0500, Mac A. Cody wrote:
Greetings,
In Steve Markgraf's slide presentation (http://people.osmocom.org/steve-m/fl2k_slides/osmo-fl2k.html), do slides 16 and 17 imply that some FL2K devices have LDO regulators while other using switching regulators? Obviously, the FL2K devices that have LDO regulators are preferred, due to fewer spurious RF emissions. How can one determine which FL2K devices have LDOs? Can an FL2K device be reworked to use LDO regulators?
Thanks,
Mac
Marcus, Thank you for your prompt, detailed, informative, and (even) entertaining reply. Many of the things you mentioned I am already aware of, but this information is definitely valuable as a reminder and for those unaware of all of the issues. You put in far more time into your reply than I would have ever hoped for and it is greatly appreciated! I'm looking at using the FL2K for amateur radio HF band applications. I want to employ a sampling rate of 120 Msps and a low-pass filter with a roll-off just above 30 MHz. I'm considering using an Odroid XU4 as an embedded processor, as it has significant processing capability and it has two USB 3.0 hosts. Of course, a suitably enhanced RTL-SDR device would be used for the SDR receiver component. Alternatively, an SDRPlay RSP1A could be used. Regards, Mac On 05/10/2018 03:48 AM, Müller, Marcus (CEL) wrote:
Hi Mac,
so first of all: any spur is only a problem if it ends up in your signal. Since we're clearly talking about devices that you can't use for operation with an antenna withou very much filtering: check whether you actually get a problem first. To be completely honest, the whole LDO vs. SMPS discussion often bares technical background, as you'll find SMPS in high-end radio receiver devices just as well. It's all about /designing/ your thing to be low noise, not about the "use an LDO instead of a switcher".
Now, Steve has offered nice figures about the spurs there, so these might actually be linked to the switcher. However, the method seems to be to first measure, then link cause to it; not the other way around. I'd argue that the device with the many spurs that, actually do look like one rectangular wave modulated another rectangular wave, was simply badly designed, probably with underdimensioned means of eliminating cross-talk between the two switchers (no idea how they relate). What confuses me is that these spurs roughly fall into a 3 MHz grid – and that's usually a bit on the high end for switching frequencies.
Another device with a switcher *might* be nicely filtered and work perfectly well. I agree, adding a switching regulater definitely adds a source of noise, but please don't assume that cheaply designed LDO systems are superior in signal quality¹; there's modern switch mode supplies that actually use spread-spectrum methods to spread out the energy they leak onto many frequencies², and others that you can synchronize e.g. to sampling clocks so that noise at least aligns and can be filtered out more easily.
The point I'm trying to make is that if these spurs are a problem to you (and I can heartily to figure on slide 17 worrying you), then you'll want to have spur measurements at different sampling rates at exactly your USB bus – in the end, the noise of a SMPS very much depends on how hard it is at work, and a stable input supply and high output current might be nicer than a dropping input and a current draw so small that forces the SMPS into discontinuous current mode.
Regarding spotting: -------------------
Switch mode supplies generally can be found by looking for (large-ish) inductors close to (large-ish) diodes, typically close to either a converter IC or in higher-current applications close to a (large-ish) discrete transistor. Do an image search for "SMD power inductor", and you'll see how these tend to look like.
Regarding remedies: -------------------
Filters, filters, filters³! You need to select the right Nyquist zone, anyway. So, pick a sampling rate range that works out for that; shifting your signal in digital domain so that it ends up where you want it after being shifted by the sample clock N times allows you to have some leeway there. Then, use whatever remaining degrees of freedom you have to pick a rate that is at a supply spur – and filter that out. Whether this is an option at all of course depends on the RF bandwidth you need.
Replacing the power supply on-board: I'm willing to say "it's possible", but I'd also say "at a time investment higher than simply buying a handful of candidates and simply sticking with one that works". Supplies typically have to be electrically well-coupled to the ground and supply lines, so if you externalize these, you'd replace the original output stage of the on-board SMPS with larger capacitors, but these typically have worse RF interference suppression properties, so you'd add smaller capacitors, but now you have a system with capacitors of different sizes and internal resistances and inherently some inductive characteristics of whatever connects the external supply to these connectors – you can certainly simply build that, and it's not that unlikely it'd work, especially if you overdimension everything a bit, but I wouldn't know how to predictably make a "first trial works" device. Note that switch mode ICs for these voltages and currents aren't necessarily solder-friendly⁴ . Rule of thumb: The smaller the package, the higher the switching frequency⁵ – and as noted above, 3 MHz would be at the higher end of the spectrum of switching frequencies⁶, but that's likely because higher switching frequency also makes the necessary inductance smaller, and hence, the inductor cheaper.
What I would do ===============
Compare a handful of dongles. Because: a) They're cheap, and time is sparse, b) can't be that bad to have spare ones lying around, for operation away from the spurs, or to give to friends who want to try that, or to honestly resell as tested to work with osmo-fl2k but replaced with a lower-noise one, c) to verify hypotheses on how to fix things, without risking to fry one of the "good ones", and c) if you can figure out how to improving the best one, maybe it becomes easy to improve the others, too. Maybe it's easier to observe an improvement in the ones that are bad.
Go and measure. That means that I'd both add appropriate output filters for both the Nyquist zone I want, and measure after that (e.g. using an RTL dongle, whose spurs I at least know), as well as trying to figure out where exactly the spurs come from – are they really on the signal lines, or are they radiated into my measurement by the shield conductor of the VGA port? When I probe around with an oscilloscope, on which lines do I see exactly these frequencies I observed?
Then, improve and adapt. If things are actually radiated by the board, proper shielding might be the simplest method to improve the situation. Else, go for easy things like soldering another (better, as in lower ESR, higher capacity?) capacitor onto the decoupling capacitors or output smoothing caps on-board⁷ first.
Best regards, Marcus
=======================================================================
¹ You can underdampen these LDOs, just as well, or underdimension them: linear supplies tend to be cheaper than SMPS for small loads, so the fact that some manufacturers use SMPSes might point out that you'd need a relatively beefy and fast LDO and thus expensive LDO to reliably supply the current needed, and there's plenty that you can mess up when you're designing an LDO system at the edge of cost efficiency
² Though that doesn't sound too desirable here
³ Imagine Ballmer going "developers!" on you here.
⁴ In highly integrated electronics, ICs with 6 pads in a package of total size ~ 1 mm × 1.3 mm would be typical if you just need a small step-down from 1.8 V to 1.2 V efficiently.
⁵ Because the higher the frequency, the less charge transfered per cycle, the lower the switched current, the smaller the switching transistor.
⁶ Please don't really infer that this means you get a chip scale package – these VGA dongles were built with cost, not size, as primary target, as you can see from the sparsely populated simple PCBs; you don't use a high-end phone-building assembly line to build 5 € VGA dongles, so you don't use <0.05 mm tolerance in placement parts.
⁷ Maybe don't take it this far: https://twitter.com/LaF0rge/status/892872883164336128
On Wed, 2018-05-09 at 22:44 -0500, Mac A. Cody wrote:
Greetings,
In Steve Markgraf's slide presentation (http://people.osmocom.org/steve-m/fl2k_slides/osmo-fl2k.html), do slides 16 and 17 imply that some FL2K devices have LDO regulators while other using switching regulators? Obviously, the FL2K devices that have LDO regulators are preferred, due to fewer spurious RF emissions. How can one determine which FL2K devices have LDOs? Can an FL2K device be reworked to use LDO regulators?
Thanks,
Mac
Marcus, Thank you for your prompt, detailed, informative, and (even) entertaining reply. Many of the things you mentioned I am already aware of, but thisinformation is definitely valuable as a reminder and for those unaware of all of the issues. You put in far more time into your reply than I would have ever hoped for and it is greatly appreciated! I'm looking at using the FL2K for amateur radio HF band applications. I want to employ a sampling rate of 120 Msps and a low-pass filter with a roll-off just above 30 MHz. I'm considering using an Odroid XU4 as an embedded processor, as it has significant processing capability and it has two USB 3.0 hosts. Of course, a suitably enhanced RTL-SDR device would be used for the SDR receiver component. Alternatively, an SDRPlay RSP1A could be used. Regards, Mac On 05/10/2018 03:48 AM, Müller, Marcus (CEL) wrote:
Hi Mac,
so first of all: any spur is only a problem if it ends up in your signal. Since we're clearly talking about devices that you can't use for operation with an antenna withou very much filtering: check whether you actually get a problem first. To be completely honest, the whole LDO vs. SMPS discussion often bares technical background, as you'll find SMPS in high-end radio receiver devices just as well. It's all about /designing/ your thing to be low noise, not about the "use an LDO instead of a switcher".
Now, Steve has offered nice figures about the spurs there, so these might actually be linked to the switcher. However, the method seems to be to first measure, then link cause to it; not the other way around. I'd argue that the device with the many spurs that, actually do look like one rectangular wave modulated another rectangular wave, was simply badly designed, probably with underdimensioned means of eliminating cross-talk between the two switchers (no idea how they relate). What confuses me is that these spurs roughly fall into a 3 MHz grid – and that's usually a bit on the high end for switching frequencies.
Another device with a switcher *might* be nicely filtered and work perfectly well. I agree, adding a switching regulater definitely adds a source of noise, but please don't assume that cheaply designed LDO systems are superior in signal quality¹; there's modern switch mode supplies that actually use spread-spectrum methods to spread out the energy they leak onto many frequencies², and others that you can synchronize e.g. to sampling clocks so that noise at least aligns and can be filtered out more easily.
The point I'm trying to make is that if these spurs are a problem to you (and I can heartily to figure on slide 17 worrying you), then you'll want to have spur measurements at different sampling rates at exactly your USB bus – in the end, the noise of a SMPS very much depends on how hard it is at work, and a stable input supply and high output current might be nicer than a dropping input and a current draw so small that forces the SMPS into discontinuous current mode.
Regarding spotting: -------------------
Switch mode supplies generally can be found by looking for (large-ish) inductors close to (large-ish) diodes, typically close to either a converter IC or in higher-current applications close to a (large-ish) discrete transistor. Do an image search for "SMD power inductor", and you'll see how these tend to look like.
Regarding remedies: -------------------
Filters, filters, filters³! You need to select the right Nyquist zone, anyway. So, pick a sampling rate range that works out for that; shifting your signal in digital domain so that it ends up where you want it after being shifted by the sample clock N times allows you to have some leeway there. Then, use whatever remaining degrees of freedom you have to pick a rate that is at a supply spur – and filter that out. Whether this is an option at all of course depends on the RF bandwidth you need.
Replacing the power supply on-board: I'm willing to say "it's possible", but I'd also say "at a time investment higher than simply buying a handful of candidates and simply sticking with one that works". Supplies typically have to be electrically well-coupled to the ground and supply lines, so if you externalize these, you'd replace the original output stage of the on-board SMPS with larger capacitors, but these typically have worse RF interference suppression properties, so you'd add smaller capacitors, but now you have a system with capacitors of different sizes and internal resistances and inherently some inductive characteristics of whatever connects the external supply to these connectors – you can certainly simply build that, and it's not that unlikely it'd work, especially if you overdimension everything a bit, but I wouldn't know how to predictably make a "first trial works" device. Note that switch mode ICs for these voltages and currents aren't necessarily solder-friendly⁴ . Rule of thumb: The smaller the package, the higher the switching frequency⁵ – and as noted above, 3 MHz would be at the higher end of the spectrum of switching frequencies⁶, but that's likely because higher switching frequency also makes the necessary inductance smaller, and hence, the inductor cheaper.
What I would do ===============
Compare a handful of dongles. Because: a) They're cheap, and time is sparse, b) can't be that bad to have spare ones lying around, for operation away from the spurs, or to give to friends who want to try that, or to honestly resell as tested to work with osmo-fl2k but replaced with a lower-noise one, c) to verify hypotheses on how to fix things, without risking to fry one of the "good ones", and c) if you can figure out how to improving the best one, maybe it becomes easy to improve the others, too. Maybe it's easier to observe an improvement in the ones that are bad.
Go and measure. That means that I'd both add appropriate output filters for both the Nyquist zone I want, and measure after that (e.g. using an RTL dongle, whose spurs I at least know), as well as trying to figure out where exactly the spurs come from – are they really on the signal lines, or are they radiated into my measurement by the shield conductor of the VGA port? When I probe around with an oscilloscope, on which lines do I see exactly these frequencies I observed?
Then, improve and adapt. If things are actually radiated by the board, proper shielding might be the simplest method to improve the situation. Else, go for easy things like soldering another (better, as in lower ESR, higher capacity?) capacitor onto the decoupling capacitors or output smoothing caps on-board⁷ first.
Best regards, Marcus
=======================================================================
¹ You can underdampen these LDOs, just as well, or underdimension them: linear supplies tend to be cheaper than SMPS for small loads, so the fact that some manufacturers use SMPSes might point out that you'd need a relatively beefy and fast LDO and thus expensive LDO to reliably supply the current needed, and there's plenty that you can mess up when you're designing an LDO system at the edge of cost efficiency
² Though that doesn't sound too desirable here
³ Imagine Ballmer going "developers!" on you here.
⁴ In highly integrated electronics, ICs with 6 pads in a package of total size ~ 1 mm × 1.3 mm would be typical if you just need a small step-down from 1.8 V to 1.2 V efficiently.
⁵ Because the higher the frequency, the less charge transfered per cycle, the lower the switched current, the smaller the switching transistor.
⁶ Please don't really infer that this means you get a chip scale package – these VGA dongles were built with cost, not size, as primary target, as you can see from the sparsely populated simple PCBs; you don't use a high-end phone-building assembly line to build 5 € VGA dongles, so you don't use <0.05 mm tolerance in placement parts.
⁷ Maybe don't take it this far: https://twitter.com/LaF0rge/status/892872883164336128
On Wed, 2018-05-09 at 22:44 -0500, Mac A. Cody wrote:
Greetings,
In Steve Markgraf's slide presentation (http://people.osmocom.org/steve-m/fl2k_slides/osmo-fl2k.html), do slides 16 and 17 imply that some FL2K devices have LDO regulators while other using switching regulators? Obviously, the FL2K devices that have LDO regulators are preferred, due to fewer spurious RF emissions. How can one determine which FL2K devices have LDOs? Can an FL2K device be reworked to use LDO regulators?
Thanks,
Mac
participants (2)
-
Mac A. Cody -
Müller, Marcus (CEL)