|
|
| Elliott Sound Products | Project 245 |
 Main Index
Projects Index
|
PCBs are available for this project. Click here for details.
Project 198 (MOSFET relay) has been quite popular, as it's pretty much the only one of its kind that's readily available. It's capable of switching DC at any current and voltage within the ratings of the MOSFETs you use, and I've not yet been able to destroy one (and I have tried!). However, the gate driver IC (Si8752) is only available from one manufacturer, Silicon Labs (aka Skyworks Solutions Inc.).
Thanks to a nice man at Texas Instruments, I can now offer an alternative. There's nothing wrong with the original, but the IC is 'single sourced', and if it becomes difficult to get (as was the case during the Covid outbreak), no-one can build the circuit. The TPSI3052 is actually better as a MOSFET driver, as it's capable of high-speed switching, something the Si8751/ Si8752 ICs aren't so good at.
We don't need super-fast switching for most things, and especially when disconnecting a faulty amplifier from your speakers, but the speed does provide other opportunities. This project is designed to be a MOSFET relay though, but all operational modes are supported so it's very flexible.
Having purchased an evaluation module (EVM) to test the capabilities of the TPSI3052, I can confirm that it is a very viable alternative. The IC is currently available from only a couple of suppliers - this will undoubtedly improve though.
Naturally, it has a completely different pinout and it works differently, but other than the requirement for a different PCB, which will be available 'shortly' (this is currently undefined). The IC can be driven in two modes - 3-wire or 2-wire. 3-wire mode allows for very fast on/ off times, but as a relay, 2-wire mode is more than adequate. When used in 3-wire mode, an 'always on' 5V supply has to be provided, which may be a nuisance. There's another mode that I found during tests, which I call 'modified 2-wire', and this is the best compromise for a general-purpose MOSFET relay.
The IC is certified according to DIN EN IEC 60747-17 (VDE 0884-17), and has a peak isolation voltage of 5kV. It is fully rated for switching 230V AC mains, and the wide package ensures very acceptable creepage and clearance distances. All-in-all, this is an excellent IC, offering good flexibility and exemplary performance.
The circuitry is a little more complex than for the Si8752, but not to the point where it's over the top. There are two caps on the secondary side that are used as part of a charge-pump used to ensure a good DC gate drive level, and its switching speed is blindingly fast if you don't consider propagation delay. The delay between applying +5V to the 'EN' (enable) pin is about 6ms, as it takes that long for the gate voltage to have reached the minimum allowed. The IC doesn't apply gate voltage until the available voltage is 12V, and then the switching speed is extremely fast. When the 'EN' pin's voltage is removed, turn-off is initiated as soon as the IC detects that the voltage has fallen below ~4.5V. Turn-off is also very fast, minimising MOSFET dissipation during switching.
The coupling between the low-voltage (control) side and MOSFET gate drive is done capacitively with the Si8752, but the TPSI3052 uses a transformer driven with an internally generated 89MHz (typical) pulse waveform. Everything is within the IC, including the transformer. In a minimalist circuit, you only need to provide three capacitors and one resistor, and this is the way I tested it. The evaluation board could be used, but it's a fairly expensive option - over AU$100.00, and it has provision for every option that can be used. It includes a pair of SiC (silicon carbide) MOSFETs, but they are an expense that's not warranted for most applications.
When used in 2-wire mode, the output isn't switched until the gate drive voltage has reached the design level, typically 12V. After 5V is applied to the 'EN' (enable) input, it takes about 6ms for the MOSFET's gate drive to be switched on. This can be reduced to ~3.6ms by using a higher voltage (e.g. 12V), but I found that this may create a narrow 'glitch' when the power is removed. This does not occur with the suggested resistor values (see below for more info). The off-time is less than 300μs - I measured 220μs. For a MOSFET relay, this is fine, and while it can be improved by using 3-wire mode, this isn't required for a simple SSR.
The datasheet is in the general style we've come to expect of late. There's every piece of information you're ever likely to need, but it's organised in a way that I find rather unhelpful. There are 45 pages, and while it does show specific examples of 2 and 3-wire modes of operation, there are still many things you need to search for. For example, there are two capacitors used for the output (gate drive) section, and these are determined by a formula that does explain it, but not in a way that hobbyists (in particular) will enjoy much.
Fortunately, a lot of the fine detail can be dispensed with, since we are interested in building a utilitarian device. The speed isn't especially important, provided it turns off quickly when a fault is detected. This means that the vast majority of the design details can be skipped. For a speaker protection relay, we really don't care if it takes 6 or 60ms to turn on, because P33 applies a delay anyway.
We do like that it will turn off in less than 1ms - that's much faster than an electromechanical relay, even if we use every trick in the book to speed up dropout time. Being a MOSFET relay, we know that we can break DC, because there are no contacts to arc. It's simply a matter of selecting the switching MOSFETs for low RDS(on), and ensure that their voltage rating has at least a 10% (preferably 20%) safety margin.
The circuit is configured to use the IC in 2-wire mode, because that's the easiest to implement. This also makes it a lot easier to drive, because a separate 5V power source isn't required. While the datasheet says that the 'EN' (enable) pin can be driven with up to 48V in this mode, I'm wary of that because I measured a small glitch (a momentary turn-on) with an input voltage above 5V. This isn't really a problem, but I don't like it. There was no sign of the glitch if the 'EN' pin is actively pulled low. There is no glitch with the recommended resistor value (2.2k).
The values of the capacitors CDIV1 and CDIV2 are fixed, so you don't need to worry about performing calculations. A value of 330nF (CDIV1) and 1μF (CDIV2 will be fine with any 'typical' MOSFET you're likely to use, and even if there's a slight mismatch from the theoretical requirements, this won't have a significant effect on the circuit's operation. This has been tested and verified - there is very little difference, even with a total gate capacitance up to 10nF.
There are several requirements for CDIV1 and CDIV2 if you intend to use the 5V supply available from the positive side of CDIV2. However, for a simple MOSFET relay circuit this isn't required, and values shown are recommended. These are the values used in the evaluation board.
CDIV1 and CDIV2 and CVDDF must be ceramic, with the shortest possible leads. The PCB (when available) will have provision for either SMD (1206 - 3 × 1.5mm, 0.12" × 0.06") or through-hole caps (5.08mm hole spacing). Most through-hole MLCC caps use 5.08mm hole spacing. The stray inductance between the IC pins and capacitors is minimised, needed because we're dealing with a 90MHz switching frequency. The IC is actually reasonably tolerant of a bit of stray inductance, and all that happens is there is more ripple on the gate supply.
The diode is to protect the MOSFET gates in case of a fault that may cause the maximum gate-source voltage to be exceeded. The IC has protection (via an active clamp) against voltage applied when the MOSFETs are supposed to be off, but there's no protection against a negative voltage. Rg is optional. If used it should be around 10Ω, but it's not necessary in a MOSFET relay. This is not included on the PCB.
The final circuit is shown in Fig. 2, and I added an input resistor (R1) to discharge any stray capacitance. With 2.2k, the maximum resistor dissipation is 65mW with a 12V supply, but I'd expect the maximum will be about 24V, as available from P33 This requires that R1 is increased to 3.3k (175mW). The datasheet recommends that C2 and C3 should be in a ratio of around 1:3, and these are the values used on the evaluation board. I can see no good reason to change them, as this is an arrangement that works well.
If you'd rather use 3-wire mode, Fig. 3 shows how. When the PCB is offered, both options will be available so you can decide which method you'd rather use at build-time. The +Ve supply is applied when power is on in your amplifier (or other project), and the 'In' input is asserted (+5V) to turn the MOSFETs on. In this mode, the IC can be driven at high speed (up to 50kHz is possible). However, operating at high speed requires very careful MOSFET selection, determination of CDIV1 and CDIV2 and power transfer. Operation at mains frequency is not a problem, and is unlikely to require any changes to the basic design shown.
One option that is glossed over in the datasheet is to simply tie the 'EN' and 'VDDP' pins together. I've tested this switching at up to 1kHz and it works perfectly. Turn-on and turn-off times are well below 1ms, and there are no special precautions needed. To achieve the best turn-off time, the parallel resistor (R1) should be reduced to 1k, but I've tested with 2.2k and saw no issues. 1k provides a time-constant with R1 and C1 of 100μs, ensuring a fast turn-off. Apart from the obvious requirement of ensuring that the input voltage doesn't exceed 5V, there don't appear to be any down-sides. I measured a turn-on time of 600μs, with turn-off time of <300μs, with a switching speed of 0.1Hz. I drove the input directly from my function generator, which has a 50Ω output impedance, and it had no problems driving both the 'EN' and 'VDDP' inputs in parallel.
Although the datasheet doesn't go into any details for this mode of operation, IMO it's ideal. It provides performance that's exemplary in all respects. The current needed is minimal, and it only needs a 5.1V zener diode to regulate the input voltage. An external resistor is used to provide a suitable working current from the output of P33. Based on my tests, allowing about 50mA zener current is ideal. When the switching frequency is increased, the two caps (CDIV1 and CDIV2) will stay charged because the only current drawn is to the gates of the MOSFETs. However, if you wanted to use the circuit as a dimmer (or other variable duty-cycle switch), the two inputs must be separated (e.g. as a phase cut mains dimmer, leading or trailing-edge).
Note that the scope captures have been edited to place two captures on the one image, and reduce the height where possible. The traces themselves are from the scope, and were not edited or changed at all.
The scope capture (a composite of the two waveforms) shows the turn-on and turn-off response, using the Fig. 3 circuit with the jumper installed. For want of better terminology, I'll call this 'modified 2-wire mode'. The input is from my function generator, and the slow rise and fall times are due in part to the 100nF cap from VDDP to ground. The input signal was a 1Hz squarewave, set to deliver from 0-6V. The voltage can be seen to rise to 6V at the beginning of the turn-on phase, but it quickly falls back to 5V (about 400μs). 600μs after the input voltage goes high, the gate drive rises to 16V, with a small slope because it turns on at 13V, before the voltage has reached its maximum of 16V (the datasheet says from 13.9V to 16.2V, with a 'typical' figure of 15V).
Turn-off is a lot faster, and the gate drive goes low less than 300μs after the input voltage is removed. Note the rise and fall times of the gate-drive signal - according to the datasheet, both are much less than 10ns. The 'blurb' says 6ns rise time and 5ns fall time. I can't confirm that as my scope isn't fast enough! However, the speed is apparent from the trace - it's vertical in both cases.
I mentioned the glitch, so I need to show it so you can see what it looks like. It occurs about 40μs after the power to the 'EN' pin is disconnected. There's no active pull-down circuit, just the simple disconnection of the 12V supply. You can see some 'disturbance' as contact was broken from the supply to the 'EN' pin. The glitch is about 1μs wide, and while it will do no harm to anything, it really shouldn't be there. This was taken from the evaluation module, so we have to assume that everything is as it should be.
Some further investigation showed that the glitch only shows up under some conditions. When I ran my initial tests, I used a 5.9k resistor in parallel with the 'EN' pin. This wasn't selected, but was the first to hand, and it provided a convenient point to connect the input. Had I used a 2.2k resistor, I would never have seen the glitch! Testing with 2.2k reveals no glitch, so I ran tests and found that the threshold is around 3.6kΩ. Above that you'll likely see the glitch, and it is delayed as the resistor is increased. The lesson to be taken from this is to use the 2.2k resistor suggested, which can be increased to 3.3k for a 24V 2-wire input. If you wish, 3.3k can be used regardless of the input voltage, but if you did want to use a 48V input, it will dissipate 700mW, and a 1W resistor is called for.
The final set of graphs shows the 2-wire switch on/off performance when driven directly from my function generator. It was set to output 0 to +10V, connected directly to the 'EN' pin, with the 'VDDP' input left floating. Turn-on is leisurely, at just over 3ms (which isn't a problem for a MOSFET relay), and turn-off takes about 100μs. This is more than fast enough for any MOSFET relay, but to achieve that the 'EN' pin must be actively pulled low.
Overall, my preferred method for speaker protection is the 2-wire connection. Turn-on is slow, but turn-off is fast, and there is minimal additional circuitry needed to get a very good result. It's also the lowest current draw, with the IC drawing ~4.7mA. R1 pulls an additional 5.45mA. D1 must not be installed, and operation at up to 24V is not a problem. The datasheet says 48V, but that requires a compromise for R1.
Interestingly, 'modified 2-wire' mode is the one technique that isn't really discussed in the datasheet or EVM documentation, yet its performance is ideal for simple MOSFET relay projects that require switching with less than 1ms on/ off times. 2-wire mode is the fastest of all.
I haven't shown the switching waveforms for 3-wire mode, because there's little point. The rising/ falling propagation delays are typically 3μs and 2.5μs respectively, and while I did confirm that, it doesn't make for an interesting display.
Unfortunately, we don't get to see an internal schematic that would allow detailed analysis. This isn't a surprise, as I'd expect it to be fairly complex overall. There's a lot going on inside the IC, including the transformer driver, the transformer itself (I think it's amazing that TI managed to integrate that!) and the modulator that passes the status of the 'EN' pin to the output. The secondary side is bound to be no less complex. Overall, it's very had not to be impressed by the IC and its internal structures. Given that the isolation barrier has a claimed lifetime of over 1,000 years at 1kV RMS, what's not to like?
One thing that needs to be mentioned is the apparent lack of protection for very fast risetimes across the MOSFETs. This was handled in the SiSi8751/2 ICs with 'Miller' capacitors, but these aren't used with the TPSI3052. Instead, there's a clever active clamp between the gate and source terminals, so if anything tries to elevate the gate voltage (when it's supposed to be off) it's clamped to less than 2V (datasheet figure - I measured 0.8V). This works even when the input side has no power! Because the gates are clamped to a voltage below the conduction threshold, a high-speed transient can't turn on either MOSFET, and no Miller caps are required. This active clamp affords protection, but D2 is still required to obtain negative protection.
The next decision is to select the MOSFETs. In general, those I recommend for the P198 board will be quite acceptable here as well. The complete details are available in the construction article, available to those who have purchased ESP boards. The following section describes the criteria you need to be aware of when choosing MOSFETs. Power dissipation with normal loading is obviously a major consideration, as no-one will want to add a heatsink.
There are countless MOSFETs that will satisfy the needs for a MOSFET relay for a speaker protection system, and (at least up to a point) we can be reasonably certain that the power dissipated in the MOSFET switches will be such that no heatsink is needed. Allowing for a 10dB peak/ average ratio for 'typical' music, we know that the average current will be around 3.16A. If the supply voltage is (say) 56V as obtained from a 40V transformer secondary, the maximum current into a 4Ω load is 14A (10A RMS). 10dB below that is 3.16A, so a MOSFET with 44mΩ RDSDS(on) (e.g. an IRF540N) will dissipate an average of only 440mW at full (unclipped) volume.
That will rise to 8.6W if there's a fault that puts 56V across the load, but that will only last for a few milliseconds before P33 turns off the MOSFETs and disconnects the speakers. There are many MOSFETs with an RDS (on) of less than 10mΩ, with quite a few rated for less than 3mΩ. As the voltage (VDS) is increased, expect to see a higher RDS(on) and a higher price.
There's an advantage to keeping RDS(on) low - lower distortion. If the voltage across a component is minimal (compared to the overall voltage) then the distortion it can contribute is also minimal. For example, if we have a 42V RMS sinewave powering a 4Ω speaker at full voltage (450W), the current is 10A RMS. Each IRF540N MOSFET will have a voltage of about 442mV (RMS) across it, and will dissipate just over 4.4W. This will only ever be for transients if the amp is kept below clipping, and the average voltage across the MOSFETs will be around 140mV each (less than 0.5W dissipation).
With a 'perfect' amplifier (zero THD), the full power distortion (simulated) across the two MOSFETs is about 0.046%, and the distortion across the load is only 0.001%. The distortion across the load is directly proportional to the voltage across the load divided by the voltage across the MOSFETs. In this case, the ratio of load voltage to MOSFET voltage is 47.5:1, so the MOSFET distortion contribution is reduced by (roughly) the same ratio. It's not quite that simple, but it's a reasonable approximation.
If the RDS(on) is halved, the MOSFET distortion falls to 0.015%, and the distortion across the load falls to 0.00016%. So, it's important to minimise RDS(on), not only to keep dissipation low, but also to minimise any distortion introduced by the MOSFETs. It's worth noting that I had to reduce the gate drive voltage to 6V to achieve the figures quoted, so reality will be significantly better with a 15V gate drive.
So, obtaining MOSFETs with a low RDS(on) is well worth the bit extra you may have to pay for them. You don't need anything exotic - using SiC MOSFETs will cost more but for no real benefit. Yes, you can boast that your circuit uses the 'latest and greatest' (although that title probably belongs to GaN - gallium nitride). Like SiC, GaN is intended for very fast switching at high voltages (which we don't need).
Even 'pedestrian' devices like the venerable IRF540N are quite usable, and are suitable for amplifier supply voltages of up to ±50V. At those power levels, it would be wise to use something different though. You can also use paralleled MOSFETs for each side of the switch. One thing that works against MOSFETs is the fact that RDS(on) increases with increasing temperature, and while this can assist with current sharing, it means that they must be kept cool to prevent thermal runaway. It's possible to include thermal sensing so that the MOSFETs are turned off if they get too hot, but that just adds another layer of needless complication.
I've always been an advocate for the adage that "the fewer the moving parts, the less there is to go wrong". These parts may not be moving, but every added component is something else that can fail.
The SiC MOSFETs included on the evaluation board are UJ4C075060K3S, N Channel, 28 A, 750 V, 58mΩ in a TO-247 case, rated for 155W dissipation. There's no doubt that these are impressive devices, but they cost almost AU$16.00 each. I really like them, but there's no need for this degree of sophistication for a speaker protection relay. The RDS(on) value is a bit higher than I'd like - ideally you'd use something with less than 20mΩ to minimise both power dissipation and distortion. Suitable devices are available for under AU$4.00 each.
Note: When a speaker is suddenly disconnected due to DC, any inductance will create a back-EMF, proportional to the inductance and current. This can reach a very high voltage that may cause avalanche breakdown in one of the MOSFETs (polarity dependent). The MOSFETs used should provide a guaranteed avalanche rating of not less than 250mJ (millijoules). To give you an idea, if the MOSFET is subjected to an avalanche of 100mJ for 100μs, that's the equivalent of 1kW across the 100μs period.
The avalanche current is limited by the DC resistance of the load and speaker leads. Very few MOSFETs will be found wanting in practice, but if you're paranoid you may wish to add a TVS diode or MOV across the speaker terminals. Either must have a breakdown voltage that's greater than the supply voltage used by the amplifier. Ideally, the breakdown voltage will be less than the rated drain-source voltage of the MOSFETs, but this implies an accurate rating (neither device has great precision).
The IC is only 8 pins, but they are all dedicated to a particular function. The pin numbers, abbreviations, type (input, output, power, ground) and function are shown in the following table. For the standard connection as a MOSFET relay, VDDP is bypassed to ground, but is otherwise not used. The power transfer is programmed by RPXFR, and Table 2 shows the datasheet recommended values.
Pin Abbr. In/Out Function 1 EN I Active high driver enable 2 PXFR I Power transfer 3 VDDP P Power supply for primary side 4 VSSP GND Ground supply for primary side 5 VSSS GND Ground supply for secondary side 6 VDDM P Generated mid supply 7 VDDH P Generated high supply 8 VDRV O MOSFET gate drive output Table 1 - Pin Designations
Power transfer can be adjusted by selecting one of seven power level settings using an external resistor from the PXFR pin to VSSP. In three-wire mode, a given resistor setting sets the duty cycle of the power converter (see Table 2) and hence the amount of power transferred. In two-wire mode, a given resistor setting adjusts the current limit of the EN pin and hence the amount of power transferred. The output ('VDRV') can source an instantaneous gate current of 1.5A and sink 3A, ensuring fast on and off times for the MOSFETs.
RPXFR Duty Current Comments 7.32k 13.3% 1.9mA 9.09k 26.7% 2.8mA 11k 40.0% 3.7mA 12.7k 53.3% 4.5mA 12k is the suggested value 14.7k 66.7% 5.2mA 16.5k 80.0% 6.0mA 20k 93.3% 6.7mA Table 2 - Power Level Settings
(1) Standard resistor (EIA E96), 1% tolerance, nominal value.
(2) RPXFR = 100k or RPXFR = 1k sets the duty cycle of the power converter to 13.3%.
The device supports seven fixed power transfer settings, by selection of a corresponding RPXFR value. Selecting a given power transfer setting adjusts the duty cycle of the power converter and hence the amount of power transferred. Higher power transfer settings leads to an increased duty cycle of the power converter leading to increased power transfer and consumption. During power up, the power transfer setting is determined and remains fixed at that setting until VDDP power cycles. It's not helpful that only E96 values are shown, as most people will prefer to use E12 values if possible. 12k is a standard E12 value and is suggested.
The limit of 4.5mA doesn't sound like much (~2.35mA at the output), but that will charge the two caps on the secondary side within a few milliseconds. That might sound 'slow', but it's faster than an electromechanical relay (EMR), and the output to the MOSFETs doesn't switch until the required voltage has been reached.
Now, while TI goes to all the trouble of providing the data shown in the tables, I tested a number of possibilities, and saw virtually no difference whatsoever. If RPXFR is open there's definitely a slow-down, but with 12k (or several other values) the gate voltage in 2-wire mode took 6ms to assert after the 'EN' pin was taken high (+5V). It was faster with 12V (about 4ms), but that's not much of an improvement. As it turns out, 6ms is a perfectly reasonable time for a MOSFET relay to activate. Deactivation is a lot faster, and the gate voltage was switched off in about 200μs. Note that these times are 'propagation delays', not the rise and fall time of the gate voltage. For all intents and purposes that's instantaneous (less than 10ns for both rise and fall).
The difference between the specifications and the results I obtained are due to the use of 5V input, rather than 6.5 as suggested in the datasheet. If the 'EN' pin is driven from 6.5V or more, most of the datasheet specs will be met, plus the one thing that isn't mentioned - the very narrow gate voltage 'glitch' (a short pulse ~20μs after the gate voltage goes low. This won't cause any problems for any 'normal' application, but it was unexpected. It's also possible that the IC on the demonstration board is an early version, but this can't be confirmed one way or another.
The datasheet explains that the TPSI5052 can also be used to drive SCRs, and it supports a 'momentary' output pulse (2.5μs) for this. This isn't useful for our application, but someone might find it be handy for alterative output devices. As noted earlier, the datasheet is very comprehensive (45 pages of it), but the organisation leaves a great deal to be desired. It's often necessary to jump 20 pages or more to find the details for operating modes, and the one I found to be the most useful isn't covered in any detail at all.
The circuit described here will almost certainly get a PCB for sale, but the timing is unknown at present. I'll probably wait until the IC is more readily available, as it's currently not especially common, being fairly new. This circuit is not intended as a replacement for Project 198, but rather an alternative. It's now fairly common for various ICs to become unavailable for any number of reasons, so having a back-up plan is a good idea.
Should you build the circuit, I'd ensure that I had at least a couple of spare ICs, kept in an antistatic bag and cable-tied inside the amplifier. Unfortunately, one never knows when an IC will be deemed 'obsolete', and having spares may save you from grief in a few years when you can no longer get a replacement.
The TPSI3052 is a very capable IC, and I recommend 'modified 2-wire mode' as it requires no changes to the drive circuit, other than limiting the supply voltage. As noted above, the 'EN' pin can be driven with up to 48V in 2-wire mode, and if you don't mind the very small gate drive glitch when the input voltage is removed, it can be powered directly from the output of P33. The current drawn when it's in operation is minimal, so there's no chance of the P33 output being overloaded.
If you'd rather use 3-wire mode, both the main supply (VDDP) and 'EN' pins are limited to 5V, so an external regulator is needed. In 3-wire mode there's also a recommended power-on sequence, and you must ensure that VDDP is provided (and steady) before the 'EN' pin is driven high. This isn't hard to achieve, but the 5V regulator is an added nuisance.
At the time of writing, the cost is around AU$11 for the TPSI3052, vs. about AU$5 for the Si8752 (both from Mouser, prices subject to change).
Note that this is a preliminary release of the project. Since it's (mainly) based on the evaluation board, I know that there are some differences between this and the final version, but I didn't get any surprises. The project PCB will be far simpler than the evaluation board, but will still include the functionality described in the TPSI3052 datasheet. I have a working prototype, which is a kluge, using a pair of SMD to DIP-8 adapters that I fully expected to have problems (two were needed because of the wide IC body). There were none, despite far longer PCB traces than I would consider acceptable, and everything works perfectly. PCB details will be included here when I have prototypes made.
To be continued ...
1 Project 198 MOSFET relay - ESP
2 TPSI3052 datasheet
3 TPSI3052Q1EVM EVM (evaluation module) datasheet
| Main Index
Projects Index
|
