Analog Voice over Power Line communication
The goal of the following project circuit is to demonstrate the feasibility of audio communication at line levels, with emphasis on voice (male and female) over low voltage (230V line to neutral) power lines.
This project has close to zero usefulness in our current age, besides design practice, for the TX part. It could have some niche use as musical / artistic performance device, or in a recording studio.
We will focus first on the RX device. Since such a device does not inject EMI on power lines, it is not a big no no in terms of use over utility mains. a derivative – and more industrially useful – project could be done on that basis to perform noise and harmonics analysis of the power line environment on the cheap.
Disclaimer
Such a device, specifically the TX part, if constructed, would not be FCC compliant or whatever normative regulation is in place in your country. As such, This article describes a theoretical device and any prototype derived from it should not be coupled to the mains, as it can disrupt the proper function of power meters, inject unwarranted noise on power lines and be hazardous to operate if proper precautions are not taken. Dangerous voltages are present in the device which can cause bodily harm or even death.
Proper handling and safety precautions when designing, prototyping and operating the device are required.
Any practical realization and testing should be done on an isolated power network (on generator power or inverter power). We are not responsible for generator or inverter failure or any other damage or unintended consequences arising from the realization of this project.
Additional disclaimer : EMI mitigation
If coupled to the mains, due to the relative proximity of the voice spectrum to the 50Hz mains frequency, a companion LP line filter device (a ‘reactor’ in electrician speak) at the point where we want to stop audio propagation would be required, and would need large inductors and capacitors. It would, however, add a low impedance path at audio frequencies, and that would absorb part of the energy that is supposed to be received on the RX end, lowering overall SNR.
In the absence of such a filter, the distribution panel and power meter would be exposed to electrical noise in the audio spectrum, which could prevent proper operation of some safety devices.
In case of testing on an isolated network (for instance, on a gasoline or diesel generator), the various filtering stages in the device exhibit a high Q factor and generator frequency stability is paramount in order for the mains hum to be adequately filtered. An inverter generator or an offline inverter test bench is preferable. The following circuit is designed for 50 Hz operation. 60 Hz operation would require offsetting the corner frequency of all filter stages.
Modelling AC mains
We will model an AC mains of 230V Line to Neutral, 50 Hz, and take into account just one phase. A crude model of a local utility 11 kV/ 230V transformer is provided.
Safety
The audio path should be galvanically isolated from mains. For this purpose a 600:600 ohm EI-14 audio transformer is used, and provides adequate bandpass on the audio band. However, such a transformer is limited to 75mW RMS power and has low reactance at 50 Hz. Connecting the primary directly to line and neutral would destroy it immediately. Such a transformer has only 140 Ohms DC resistance on primary and secondary windings and barely much reactance, since its primary inductance (measured at an excitation signal of 1kHz – as an audio standard – is only 230mH). Since magnetizing current is inversely proportional to inductance, it would reach an unsafe level, overheat the transformer windings and melt them.
The solution is to high pass (AC couple) the primary to reduce voltage such as the power dissipated from magnetizing current stays well below 75mW power RMS. This portion of the circuit is fundamentally similar to what is used for PLC (power line communication) : a high pass filter (C3 + R4) plus a pulse transformer. (L1 + L3 // L2 + L4). This forms our first filtering stage.
In our case, instead of a fc at several hundred kHz, it is closer to the power line first odd harmonic (150 Hz) and uses an audio transformer instead of a very low inductance pulse transformer.
Another possible option to reduce the 50 Hz signal at the signal transformer input is to use a capacitive voltage transformer (CVT) which in our case would be a simple capacitive voltage divider. The frequency response is slightly different with a theoretical 6dB resonance boost around 1kHz with the CVT compared to the high pass filter, but with a steeper phase shift due to the resonance, which could give a less natural sound.
Since we are working on audio signals, it could be convenient to make the whole device work on a balanced line. Low cost EI14 600 ohms audio transformers have only two output terminals, so, to emulate a center-tapped transformer, we have to resort to two matched transformers with the secondary wired in series, and we can wire the primary in parallel after the line LP filter. This would get use a 1:2 (pri:sec) voltage ratio and a 40dB/dec rolloff.
The center point of our two secondaries wired in series (L2 + L4) would be our audio ground reference, and also the power ground. it would be bounded to the center tap of a two rail +/-15V power transformer. This other transformer will power op-amps after a rectifier bridge and two 12V regulators (LM7812 + LM7912). The power providing transformer is modeled by (L9 + L7 // L8 + L10). Note that the supplied inductances inductance should be modeled according to the measurement of the device you will use, and that they may skew the operation of the subsequent filters. Which means you will need to adapt the model downstream. In practice however, tuning would be provided by potentiometers to adapt to changing conditions and different upstream component parameters.
Finally the TX device – not modeled – is replaced by a voltage source (V1) stimulated by an audio signal (wave file) containing a voice sample, and with the output characteristics of an audio amplifier : output stage at 8 ohms impedance. At this node the signal is mixed with the mains power (‘line’ node, from the 11kV/230V transformer network)
The output of the first HP (out1/out1b) stage after the audio transformer (L1 + L3 // L2 + L4) exhibits max voltage gain at around 1Khz, at +6dB (due to transformer configuration), a 40dB/dec rolloff high pass, with an attenuation of close to 56dB at 50Hz.
Just downstream of the differential audio line (out1/out1b) transformer output, we add a fully differential buffer op amp so as not to load substantially the audio transformer. a LTC1992 would fit conveniently.
Following that, it would be useful to obtain more 50 Hz rejection by adding a high Q band stop filter tuned at 50 Hz. We will use two filters for each leg of the balanced line, referenced to audio ground. This gets us an additional attenuation of 10.7dB at 50 Hz.
With the following parameters declarations for the filter parameters : (Parametrization permits stepping the filter parameters, to find the best theoretical filter parameters values)
Not all fully differential op amps are capable of performing filtering without ringing artefacts, and that was seen in the simulation. For this purpose, we used two single ended op amps at each stage except for the buffer op amp after the transformer. Calibration should be performed so as to obtain symmetrical and maximum attenuation on both legs, which would require matched op amps and 1% tolerance components, particularly for the band pass and band stop filters.
The other option is to use fixed resistors with the double of the required resistance value, and add a trim potentiometer in parallel that is then fine tuned so that each signal leg performs symmetrically. A // resistor configuration minimises noise during trim pot operation, as wipers may not always contact fully, and particularly during tuning.
For the band stop stage, trim pots would be used across R2,R3,R10 and maybe R11 and the respective resistors on the other leg.
Besides the fully differential buffer, the remaining op amps (for the remaining stages) are LT1037 and performed adequately in the simulation.
Pushing the 50Hz attenuation further.
We will use a complementary circuit that uses a standard 2VA +-15V AC/DC transformer to power up opamps (after a regulator) and to perform an additional function. It can be used, not only to power the device components, but also to provide a 50 Hz signal at a lower voltage than mains, that can be further attenuated with a flat response by a simple voltage divider for it to stay well below the op_amp rail levels.
This signal is not affected by the first pass HP stage. (C3 + R4). It will then be used to obtain better attenuation. However care should be taken to minimise distortion and noise induced from the rectifier diode commutation, by adding a snubber.
We would then perform the opposite of the band stop stage on the first audio path, and add a high Q band pass filter tuned at 50 Hz.
This filtered 50 Hz signal would then be passed to an all pass filter (phase adjust) to make it 180° out-of-phase with our audio signal from the first signal path, and then mixed with a weighted inverting summing amplifier to cancel the 50 Hz signal from our audio line even more. Proper adjustment would use two sets of (coarse/fine) potentiometers, one for the phase and the other for the gain. each potentiomerer could be ganged so as to adjust both legs of the phase adjust and mixer.
A theoretical total attenuation of close to 108dB is obtained at 50 Hz.
Stability of the 50 Hz cancellation.
Since the two mixed signals to achieve active cancellation take two distinct audio paths with different filtering characteristics, the 50 Hz component will be out of phase with respect to each other. The 50 Hz band pass signal path should be brought 180° out of phase before the summing amplifier, and its amplitude should be equal. For this purpose, an all-pass filter is used and tuned to add some phase shift to obtain a 180° phase shift between the two signals.
The following factors have to be taken into account for the cancellation not to drift significantly :
- Frequency stability of the 50 Hz grid : Usually free-running at +- 10mHz around 50 Hz. Above that level, grid primary controls (generators controllers) kick-in to bring back the frequency to 50 Hz, grid-level secondary controls from dispersed PMUs units (synchrophasors) are also used to bring mains frequency to stability, but rare deviations up to 0.1 Hz under heavy power use / power production unbalanced states cannot be ruled out. Since the band stop and band pass filters operate at the corner frequencies, phase shift is in the center of the transition range where d(phi)/df is maximum, and opposite sign between the bandpass and bandstop filters. A slight frequency deviation will make the phase shift drift roughly equal to $$ \phi_{drift} \approx \Delta (f)*(d(\phi_{bandpass}) + d(\phi_{bandstop}))/df $$
- Mains voltage sag and swell : Although both audio paths react to voltage gain changes linearly, if the power transformer used for the 50 Hz bandpass signal path is also used for op amp power supply through a regulator, it can affect loading of the transformer in a non linear manner. This would have a non linear effect on relative amplitude of the signals, and the gain of one leg of the summer amplifier would have to be adjusted. This effect would be more noticable on low power PCB transformers with poor voltage regulation. (where loading creates a higher voltage drop). A high VA rating power transformer with a no load over load voltage ratio close to unity is thus preferable; another option is to use a separate transformer for op-amp power supply. The input impedance of a buck converter can be described by the following formula :
- D is the duty cycle and depends on V(in)/V(out) ratio and is a quite significant term, and quadratic. Frequency effects are negligible for mHz drifts of mains frequency. In our simulation, adding a LM7812 / LM7815 pair of voltage regulators for op-amp supply was sufficient to throw cancellation out of balance.
Sending an audio signal over the power line : Line driver design.
So far, we have only taken into account the receiver circuit design. Now let’s dwelve into driving the power line with an audio signal.
We know that :
- impedance seen at the subscriber level with no load drawing power is quite low, and is mainly dictated by the local 11kV/400V transformer supply energy to the utility customers, as well as any filters located in the power meters or elsewhere that absorb unwanted high frequency signals used in metering or for EMI mitigation. The fc of these filters is mainly dependent on the PLC technology and frequency bands used by power utilities companies.
- Impedance is variable according to power line loading and may include reactive components if inductive loads such as motors are used.
- Broad spectrum noise due to light dimmers, led drivers, switching power supplies and power factor correction circuits is to be expected. This can have an effect on current but on voltage as well.
- Rectifiers induce low frequency harmonics of 50 Hz
- Not all noise sources will spectrally peak in the 20-20kHz audio band.
When using off the shelf line drivers, we have mainly two options : line level driving ( -10dBV consumer or +’dBu professional). +4dBu would be preferable, but the main problem would be output impedance would be still too high to properly drive an AC power line. Line level impedance is typically comprised between 100 to 600 ohms, with the recent trend going to lower impedances. The impedance mismatch is not the only factor to take into account. Line drivers are designed to drive significantly higher load impedances. They are designed to provide a signal, not power. A low impedance load would significantly overstress the output op-amp in terms of current. <current capability of line drivers ?>
The other option being amplifier “hot” signals, such as those designed to drive speakers. Impedance levels are usually 4 to 8 ohms. Galvanic isolation through a high power 1:1 audio signal transformer is highly recomended, plus a high pass filter with the same characteristics as the filter in front of the audio RX transformer, so as not to couple the audio amplifier stage with too high levels of AC 50 Hz.
This would give a first order filter with 20dB/decade slope with a 1: 1 transformer arrangement.
Digital Filter equalization.
Once the RX signal goes through the DAC, in order to obtain a flatter response for audio signals in the audio chain, we have to apply a digital equalization : a low shelf digital filter with a slope of 60dB/decade (we have to take into account the filtering at the TX stage) with a low transition frequency at 20Hz and a high transition frequency at around 730 Hz, where the gain reaches 0dB is necessary. The previously filtered trace 50 Hz signal at the analog level would get a boost, which can be further rejected by a high Q band stop Butterworth digital filter.
Going full duplex.
It would be nice to use a single device for both RX and TX, and filter the TX signal generated at the same endpoint, leaving only the reciprocal signal heard from the remote endpoint. This is known as sidetone reduction. On our RX side a 180° out of phase copy of the signal sent to the TX driver would be combined in the RX audio path. Depending on the placement, different equalization curves would be required on the TX signal copy to account for the return signal filtered characteristics. Gain should be adjusted in tandem (at the TX send level and sidetone mixing level) through either carefully calibrated ganged potentiometers or through an AGC setup. Detecting the absence of sidetone through analog means is not an easy task. digital correlation filters on the other hand could be used.