Single Supply BTL Amplifier Speaker Protection

Project 33 ^[ 1 ] has long been used by a great many people to protect their speakers against DC created by an amplifier failure.  However, it's designed specifically for amps that have a single (ground-referenced) output, so the speaker connects between the amp output and ground.  While this covers amps that use a dual supply, it's not helpful for single supply amps operating in 'BTL' (bridge tied load) configuration.

Many of the Class-D amps available (either as kits or pre-made boards) use a single supply, so the speaker outputs are at roughly half the DC supply voltage.  This is fine for the speaker, as it 'sees' a zero net voltage with no signal, but P33 cannot be used because it expects a ground reference.  If connected to either speaker line, it will detect that as having a permanent DC fault, and the relay will never operate.

To be able to work with a single supply BTL amplifier, a DC detector circuit must be able to use a reference voltage that's equal to the quiescent (no signal) DC level from the two amplifiers.  Unfortunately, the task is made harder because the nominal DC level is not necessarily exactly half the supply voltage, and it can vary with signal level.  This makes reliable detection harder than with a 'conventional' dual supply power amp.

For the most part, the descriptions here assume the use of a Class-D amp, but the circuits described will also work with single supply Class-B BTL amplifiers.  These are (or were) common in car audio systems.

Please Note:  This project is designed to work with BTL amplifiers only.  While every care has been taken to ensure it will provide protection, there may be unforeseen faults that may confuse the detector and prevent speaker disconnection.  I've tried to cover all possibilities, but anything unforeseen is (by definition) unknown.  The most likely faults are described and the detector's behaviour investigated, but this can't cover every possible fault that may develop within an amplifier (particularly Class-D).

It is vitally important that the intending constructor understands the difficulties faced when trying to break a high DC current with a relay.  Because DC is continuous (by definition), there is no period when the voltage falls to zero, and once an arc is started it is almost impossible to stop it.  The DC resistance of the speaker voicecoil is the only thing that limits the current, and in the case of P33, the relay is generally considered to be 'sacrificial' - if the amp develops a DC fault, the relay will protect your speakers, but will almost certainly be destroyed itself in the process.  This assumes that the amplifier has rail fuses (many don't) which will finally open before the relay contacts have burnt away to nothing.

In the case of a single-supply BTL amplifier, the relay can only become open circuit, because you can't use the NC (normally closed) contact to short the speaker to ground.  This makes breaking the arc a great deal harder, which will be a major limitation with amps operating from high voltages.  Most relays have a fairly low contact voltage rating for DC - a 230V, 10A AC rated relay will be downgraded to around 30V for DC.

Any potential constructor of the circuits described here is duly warned that no system can ever be foolproof, and there is always the risk that a poorly chosen relay simply will not survive, and may arc for long enough to cause speaker damage.  It is the duty of the constructor to run proper tests with the selected relay to ensure that it can break an arc reliably if there is an amplifier DC fault.

For most medium power amps (i.e. operating from no more than 30V DC - good for around 100W into 4 ohms), a 'typical' 30V 10A relay will probably be fine.  For amps running higher supply voltages, all bets are off, and it may be necessary to use two sets of contacts in series.  You can also simply hope that the DC protection circuit never actually has to detect DC.  However, consider the fact that wishful thinking has never saved a speaker from destruction (nor anything else for that matter).

There's a great deal of further info about relays in the Relays article (read Part 1 and 2).  They are not as simple as they seem at first, and DC poses special problems that may be intractable if the voltage is high enough.

Note:  Where DC voltages greater than 30V are present, I strongly recommend the Project 198 MOSFET relay.  For high-powered systems you'll need very low R_{DS (on)} MOSFETs, and/or a heatsink for them, as they may have to dissipate more power than 'typical' Class-AB designs.  MOSFET selection is discussed in the project construction article for the circuit, but is properly selected they will handle the power, and be able to break the DC fault current.  The MOSFETs used must be rated for at least 20% greater voltage than the supply, so for a single-supply Class-D amp with 80V rails (around 400W into 8Ω), you need a minimum voltage rating of 100V, and a minimum current rating of 14A - continuous!  The worst-case current will occur is one amp fails to the positive rail and the other to ground.  80V across a 4Ω load is 20A.  No electromechanical relay can break that combination - the entire contact assembly will be burnt to a crisp!

While DC detection isn't especially difficult, the BTL configuration, static DC levels and presence of the audio signal complicate matters.  We have to be able to detect a DC offset between the two outputs, but ignore the AC component.  P33 does this by using a filter that removes the AC above a given frequency, and while the same thing can be done with a permanent DC level present, there is an inevitable delay before the DC conditions have settled to the steady-state value.

The simple transistor-based circuit used in P33 won't work very well if there is a steady DC value, even if re-designed to compensate for the offset.  This is largely due to the fact that the DC level will change depending on the supply voltage.  Obtaining the DC reference isn't too hard, but the P33 detector relies on the presence of a fixed and very stable reference (ground).  This is not available with a single supply BTL amplifier.

We also have to consider the possible fault modes.  We could have one or both outputs sitting high or low, either together or in opposite directions.  If both outputs produce DC by the same amount and in the same direction, the speaker is not affected.  This is how the amp normally operates, with (about) half voltage being normal.  Should the outputs include DC of opposite polarities, this is easily detected.  The table below shows the multiple possibilities.  'Steady' simply means the normal DC offset that the amp provides when working normally.  In reality it may not be steady in the normal sense of the word, but may move up and down with the supply voltage.

Note that there is no guarantee that any particular fault condition will send one output high and the other low (other than DC at the input of a DC-coupled amplifier), nor is it more or less likely that only one output will show a DC fault.

Anything is possible when an amplifier fails! That means that any combination of the states shown is possible, depending on the exact nature of the fault.  Provided both outputs change by the same voltage and in the same direction, there is nothing to detect and the loudspeaker is safe.  We have to look at any condition where the first output is positive or negative, and the second is either at the normal steady state voltage, or has the opposite voltage to the first output.  'First' and 'second' are interchangeable in this context.  Also needed is a reference voltage, being the average value of the two BTL amplifier outputs.  This is used by the next part of the circuit - the comparators.

A fairly easy and effective technique is to use opamps (or preferably comparator ICs) in a configuration known as a 'window comparator'.  As long as the input signal remains within the 'window' of allowable voltages, the circuit's output is off, indicating that there is no problem.  Should the input go above or below the threshold, the output is active, and can be used to release the DC protection relay.  Since the application is for a BTL amplifier, there are two outputs, and two window comparators are needed.  The detector circuit in P33 is a simplified window comparator, but it uses diodes and a transistor rather than opamps.

Figure 1 - Block Diagram Of DC Detector

The above drawing shows a simplified block diagram of the detector for one channel.  R1 and R2 derive the reference voltage directly from the amplifier outputs.  It has some filtering using C1 to ensure that high frequency switching noise doesn't affect the output.  There are two window comparators, and if either operates the relay is de-energised, disconnecting the speaker.  As shown, the comparators would operate from the main DC supply, so the system is limited to a maximum voltage of 36V DC (depending on the comparators), which gives a reference voltage of around 18V.  A regulator (not included) isn't necessary if the supply voltage is less than 36V, but is needed to limit the supply for higher supply rail voltages.

The supply voltage can cause problems.  Some BTL amps (especially high power Class-D types) operate at comparatively high voltages.  Some can operate with a supply voltage of +80V or more, enabling an output power of 375W into 8 ohms, and more with higher voltages.  Opamp based window comparators can't be run at such a high voltage unless you are willing to use specialised high voltage types (which are expensive).  As already noted, the simple transistor based window comparator used in P33 is not well suited to operation with a static DC potential.  The above arrangement can be used if the output levels are attenuated or level-shifted (more on this later).

Therefore, the overall circuit is more complex.  Just how much more complex it gets depends on other facilities that may be considered desirable (or essential).  At the very least we have to detect a DC offset between the amplifier outputs, and ideally we'll also have a mute facility that doesn't connect the speakers until a timeout period that eliminates switch-on noises and/or transients.  Most Class-D amps already provide some degree of muting, but other single supply BTL amplifiers may not.  It may also be necessary to provide muting during the power-down process, as some amps may create large transients as they switch off.

Obtaining the reference voltage is easy, and it only needs a resistor from each output, and the centre tap is the reference (with filtering as shown).  This works because the output from each amp is complementary, so when one moves towards the positive rail, the other moves by the same amount towards ground.  The net result is a DC voltage that represents the exact DC level from the amps, and it follows any DC output variations that may occur as the power supply rail voltage increases or decreases (due to loading, mains voltage changes, etc.).

Then we have to detect if one or the other amp has a DC shift that indicates a fault condition.  Don't be lulled into a false sense of security by datasheets that tell you the amp has protection.  It may well have, but it only works if the amplifiers are functioning normally.  If an output device short-circuits (the most common failure mode for semiconductors), the amplifier's protection circuits cannot correct this.  Shutting down the amplifier doesn't help - the short is still present.  Some have a 'fault' output, that will be asserted if the IC detects an abnormal condition.  In some cases, this may be enough to de-activate the speaker relay, but unless the fault pin is made available and is known to be active for abnormal DC conditions, I wouldn't count on it.  Even if it is fully functional, a major internal failure may render the entire IC inoperable and the fault output may also be disabled as a result.

The net result of all of the above is a relatively complex circuit.  It has to derive a reference voltage, monitor each amplifier output, and switch off the speakers if any voltage is outside the preset parameters.  The audio part of the signal has to be removed so that only the relative DC voltages are monitored, so low pass filters are needed to remove as much audio as possible.  P33 does all of this (and more), but it's limited to being used with speaker outputs that are ground (zero volts) referenced.

There is one fairly major issue with any DC protection circuit using a relay.  If DC is present, the relay contacts will arc when they open.  P33 grounds the normally closed contact so the fault current flows directly to ground, but this can't be done with a single supply BTL amplifier because the two outputs always have DC present.  This seriously limits the maximum DC voltage that can be broken, and it will often be necessary to use two sets of contacts in series to prevent the arc current from flowing through the speaker.  See Relays And DC above for more details.

The input filters are fairly critical.  They are responsible for removing the AC component of the amp output signal, but need to be fast enough to ensure that a DC fault is detected quickly.  This would indicate that at least a second order filter (12dB/ octave) would be needed, but P33 has shown that this is an un-necessary complication.  Yes, a second order filter will (at least in theory) allow the detector to operate a little faster than a simple first order (6dB/ octave) filter, but nearly all DC protection circuits ever used or published have a 6dB/ octave filter (including P33), and no-one has shown that anything faster is necessary.

When an amplifier is used with electronic crossovers and is powering high frequency drivers, it's a simple matter to change the filter characteristics so that the detection speed is increased.  You don't need to allow 20Hz operation for a midrange or tweeter driver (or a compression horn driver), so the filters can be set for a higher frequency than an amp driving a full range or low frequency system.  Where electrolytic caps are required, they do not need to be bipolar (non-polarised) because they have a polarising voltage from the amp outputs.

These are close to the same values suggested for P33, and they have been proven to work well with countless PCBs built.  The filter outputs then go to the window comparators.  The following section describes the window comparators, and I have found that the single transistor arrangement does work reasonably well, although it is slightly asymmetrical.  The dual comparator version is 'better', but isn't essential.  However, it is recommended in this application.  Full range means that frequencies down to 20Hz are expected, and it's important that even a full power sinewave at 20Hz does not cause the relay to activate.

You can see that the f_-3 frequencies (-3dB) are somewhat lower than you might have expected.  The reason is simple - we don't want the circuit to open the relay with normal programme material, and that will happen if the filter frequency is too high.  We will be aiming for a DC detection voltage of ±2V, so the filters have to ensure that the full power signal is properly filtered out, but will activate the DC protection relay as quickly as possible if a DC fault is present.

For the comparators, you can use opamps or transistors.  As already noted, the simple transistor version as used in P33 can be made to work well, but the comparator version is preferred.  You need two comparators for each input, wired as shown below.  There is no need for any level shifting circuits if the supply voltage is less than 30V, but for higher voltages the inputs have to be attenuated to ensure that the opamp inputs aren't damaged.  Attenuation also requires that the value of R3 and R4 have to be changed so the detection voltage is not affected.  This is described in detail further below.

Figure 2 - Window Comparator Example With DC Offset

Assume that VREF is +15V, being the expected voltage found at each amp output.  Provided the comparator input is greater than 13V and less than 17V (i.e. no potentially damaging DC present) the output of the comparator remains high and the LED is off (or the speaker relay remains energised).  When the input exceeds either threshold, the output falls and the LED is on (or the relay is de-energised).  Having already removed the AC (signal) with the filter, the circuit only reacts to DC or subsonic disturbances.

We need two of these comparators - one for each amplifier output.  They can both share the same output though, because comparators are made to allow this.  (Note that opamps can be used, but then an isolation diode is needed at each opamp output.)  With the comparators, the normal output (no DC faults) means that the outputs are off, and if either (or any with multiple ICs) turns on, the transistor switch is turned off and the relay opens.

The two window comparators can also share the reference voltage (VREF) and the resistor chain of R2-R5.  This minimises the parts count.  There are several other options that could be used for DC detection, but most become too complex and/or expensive.  I firmly adhere to the statement (allegedly) made by Albert Einstein that "everything should be as simple as possible, but not simpler" ^[ 2 ].  It can be easy to over-complicate any problem, but it can also be easy to simplify it to the point where the solution no longer works.  This is rarely helpful (other than an exercise in what not to do.) 

Something to be aware of is that even where the two amplifier outputs have a DC offset (from the normal half supply level), if it's in the same direction and the same amplitude, the detector won't operate.  For example, if a BTL amp running from a 30V supply has both outputs at (say) 3V DC or 27V DC, VREF will be at the same voltage and the detector will not see a fault condition.  The relay will close to connect the speakers.  The amp will not be working with either condition, but it can't produce a DC voltage across the speaker.  The DC detector only looks for DC across the speaker - it does not indicate that the amplifier is working normally.

I haven't included the single transistor detector here because I don't think it's good enough for the job for a single supply detector.  However, Click Here to see how it can be done.  The detection polarity is reversed (LED is on with no DC).  The finer points are up to you if you wish to go this way.  The relatively large number of parts means it will take up more PCB space than a comparator based circuit.

In general, using a power-on muting circuit is essential.  This allows the amp outputs to settle to normal steady state levels before the relay closes, but if there's a problem (such as offset DC at the outputs) the relay will remain open.  Whether you also need a power-off mute is up to the constructor - if the amp shuts down silently there's no need.  It's a bit trickier to provide power-off muting, because it's necessary to use a 'loss of AC' detection circuit.  It's not difficult, but it adds parts which take up PCB space and requires another wire from the power transformer.  Power-off muting is not included in the design shown.

In the original P33 circuit, the power-on mute is added to the comparator input.  This won't work properly here, because the reference voltage is not zero and the results may be somewhat unpredictable.  The P33's loss-of-AC detector is fairly basic, but it works very well in practice.  If you need this, you will have to work out a way to implement it yourself.

So now it's just a matter of putting all the bits and pieces together to create a complete circuit.  By necessity, each BTL power amp will use its own detector circuit, and (if there is sufficient interest) the PCB will be for a single BTL amp, so two are needed for stereo.

Even with a single module on each PCB, there are quite a few parts.  It makes no sense to use a fully discrete design because two transistors occupy as much space as a dual comparator, and many more support parts (resistors, diodes, etc.) are necessary.  With two dual comparator ICs and a relay drive transistor, we can accomplish everything needed, with the exception of a loss-of-AC detector.  At least for the time being, I'll assume that it is not required.

When everything is put together, the final circuit is shown in Figure 3.  The reference voltage and its bypass cap (C1) are included.  This provides a delay when power is applied (power-on mute).  There is also a supply bypass cap (C4) which is necessary to keep the comparators stable.  The circuit will work without changes with any voltage between 24-36V, but R10 will need to be re-calculated as described below.  You may choose to modify the voltage thresholds for the window comparators.  The maximum DC supply voltage is limited by the allowable supply voltage for the LM393 comparators (36V).

With a 30V supply and the values shown for R6 and R7 (15k), the threshold is 15V ±2V (i.e. 13V and 17V).  Changing these resistors to a lower value reduces the window voltage and vice versa.  Likewise, increasing the supply voltage increases the thresholds (e.g. ±2.34V around the 18V VREF with a 36V supply).  Reducing R6 & R7 to 12k restores the ±2V window (actually ±1.92V).  For a 24V supply, use 12k (±1.8V).  The voltages can be calculated for any supply voltage and resistor value using Ohm's law.

Figure 3 - Complete Circuit Of Single Supply BTL DC Protector

If preferred, you can use an LM339 quad comparator instead of the two dual ICs shown.  The pinout is very different of course, but it's easily adapted to the circuit.  There is a very small size advantage because the LM339 is a 14 pin IC, but it's negligible in the greater scheme of things.  Performance is unchanged, as the 339 is very similar to the 393 internally.

R10 is selected so the voltage across the relay coil is correct.  For example, a typical 12V relay (as used for P33 or P39) has a coil resistance of 270 ohms.  That means it draws around 45mA from a 12V supply.  R10 is determined by Ohm's law, so that there is 12V across the relay when Q1 is turned on.  If the supply is 30V as shown, R10 needs to be 400 ohms (use 390 ohms, 1 watt).  In most cases, it will be better to use a 24V relay, as it will draw less current and won't need as much series resistance power dissipation.  Naturally, this only applies if the supply voltage is 24V or more (a BTL amp running from a 24V supply can deliver up to 32W into 8 ohms).

The amplifier is connected with both outputs going directly to the DC protector, and the speaker connects via the normally open relay contacts on one side, and directly on the other.  The wiring is shown below to make it easy to see how everything connects together.

Figure 4 - Wiring To The BTL DC Protector

The speaker is connected via the normally open relay contacts.  As noted earlier, it's not possible to ground the relay's common pin as is done in P33, because the amplifier outputs are floating at (or near) half the supply voltage.  This means that amps using greater than 30V supplies may cause contact arcing when the relay opens, and protection is compromised.  Adding a capacitor in parallel with the contacts might help to extinguish the arc.  10µF is shown, but a larger non-polarised cap will provide improved arc quenching.  Note that when the circuit has muted the amp (because the relay contacts haven't closed), there may be some amp noise that's passed by the capacitor.

Note that two circuits are required for stereo, but you can use a single relay switching transistor, with the two relay coils wired in series.  The two comparator circuits (Left and Right) can share the power supply, but each needs its own set of filters, window comparator setting resistors and its own VREF circuit, because the two separate BTL amps may not track each other for DC offset.

As noted, the circuit as shown is designed for BTL amps operating from a single supply, typically from +24V to +36V.  Amps running higher voltages can be used, but the detector circuit must be limited to an absolute maximum of 36V with the comparators shown.  In addition, the two inputs and VREF have to be limited to no more than 18V (36V supply) by means of resistive attenuators.  For an amp operating at (say) +60V, the amp outputs will be at +30V.  Each input to the detector must be attenuated by a factor of 2 to get back to 15V, and the supply to the detector has to be regulated to 30V.

Note that Q1 (the relay switching transistor) must be rated for the full supply voltage, and you may need a Darlington device if the relay has a coil current exceeding 50mA or so.

The values for R6 and R7 are shown below, assuming a supply voltage of 30V to the detector.  You will need to re-calculate the values for other supply voltages.  Note that the values shown below assume that VREF is exactly half the detector's supply voltage.  If VREF is higher or lower, the detection thresholds will be asymmetrical.  If at all possible, try to ensure that VREF is as close to 15V as possible (assuming a 30V supply for the comparators).

The general scheme is shown below.  It's important to understand that when the outputs are attenuated, the detection thresholds are also affected.  If the detectors look at 15V ±2V (13V and 17V), the effective thresholds become 30V ±4.6V if the amp outputs are attenuated by a factor of 2 (assuming 30V DC at the amp outputs with a +60V power supply).  The dividers for the window comparators (in particular R6 and R7) should be reduced if an input attenuator is used.

Figure 5 - Attenuators For High Voltage Operation (> 36V Maximum)

The attenuators are simple resistive dividers, and in the above the audio filters and VREF circuit are included for clarity.  However, there are interactions, because we have to derive VREF from the outputs as well.  With a 2:1 attenuation using 1k resistors and everything else as shown in Figure 3, a 2V DC shift in either amplifier causes an 870mV offset at the detector inputs - not the 1V you might hope for.  This is where Table 3 comes in handy, as it's immediately apparent that R6 and R7 should be about 5k6 so a DC offset of less than 1V can be detected reliably.  If preferred, you can use zener diodes in place of RA1 and RA3 to eliminate the reduction of the sense voltage, although the improvement to the overall circuit is probably not worth the extra cost.

The resistor values need to be worked out for the actual operating voltage of the amplifier.  The 60V version shown is an example, but there are many possibilities.  Aim for DC input voltages to the detector of between 10V and 20V, with 15V being ideal if a 30V regulated supply is used as shown next.  The comparators don't really care if they are operated asymmetrically (input voltages other than 1/2 supply voltage), but adequate headroom is needed to ensure that they will reliably detect a DC fault condition.

Figure 6 - Relay & Regulator For High Voltage Operation (> 36V Maximum)

Using a DPDT relay with the contacts wired in series in strongly recommended, because few relays will be able to break more than 30V reliably.  This is a problem with any high power amplifier, and there are few alternatives if high voltage, high current DC must be interrupted.  It's more critical in a BTL amplifier because the fault current can't be shunted to ground by the relay (as is done with P33 for example).

The supply voltage to the detectors must be regulated, and using a simple zener regulator as shown is quite sufficient.  All resistor values are determined using Ohm's law, and they are easy to work out.  R10 (the relay's series resistor) also needs to be re-calculated, based on the same formula (and using the same relay) shown earlier...

R10 = ( Vcc - Vrelay ) / Irelay
R10 = ( 45 - 12 ) / 45mA = 733 Ohms (use 680 ohms, 5W)

There is no reason not to use a 24V relay if you can get one, and you'll need to re-calculate the value of R10 based on the coil resistance of the relay actually used.  There will often be a significant advantage if you use two relays (or a double-pole version) with the normally open contacts wired in series.  This improves the relay's ability to break the DC current, something that can cause serious problems if not addressed carefully.  See the Relays article (parts 1 and 2) for detailed info on the ability (or otherwise) of a relay to break a significant DC current.

R11 is the current limiting resistor for the detector's power supply regulator (zeners D2 and D3).  Use 1W zeners, which should be operated at a current of 10mA - 20mA, and R11 is set to allow for a detector current drain of 10mA.  Total current through R11 is therefore up to 30mA.  If the supply voltage is (say) 45V, R11 is calculated by...

R11 = ( Vcc - 30V ) / 30mA
R11 = ( 45 - 30 ) / 30mA = 500 Ohms (use 470 ohms, 1W)

A lower current can be used if preferred, but don't allow the zener current to fall below 10mA.  The zeners will run a little warm, as they will be dissipating up to 250mW.  This depends on the actual current drawn by the comparators - it's shown as between 0.8mA and 2.5mA in the datasheet.  Zener dissipation is reduced by using two in series, and this is the most sensible option.  A single 30V zener can be used if preferred, but it will run much hotter and will be consequently less reliable over the long term.

1 - Project 33 - ESP Speaker Protection Circuit
2 - Quote Investigator
3 - LM393 Dual Comparator Datasheet
4 - Project 198 - MOSFET relay

Main Index Projects Index