1. Introduction
Network visibility of synchrophasors (i.e., voltage, current and synchronised phase) has improved markedly in recent decades. The benefits of improved visibility are well documented, including faster outage restoration, earlier detection of potential failures, opportunities for forensic analysis of power system events and a greater capacity for renewable generation [
1,
2,
3,
4]. However, the economic justification for monitoring, and the requisite communication, is often difficult for distribution networks with low load-density or rural areas with limited infrastructure. In such cases even the most basic of monitoring is often deemed too expensive relative to the potential payback [
5]. Therefore, solutions targeting deployment on these networks should see cost as one of the key considerations.
A low-cost solution for synchrophasor and/or basic loading monitoring should overcome two main challenges. Firstly, how should voltage, current, and their synchronised phase be measured inexpensively? Secondly, how should this information be communicated from its point of measurement back to the concentrator, again, inexpensively? [
6]. In traditional schemes, each monitoring point requires at least one satellite based radio navigation system (e.g., Global Positioning System, Galileo or GLONASS), transducers, an Analog to Digital converter (ADC), a phase-locked oscillator and hardware/software capable of processing the incoming ADC data and compute the synchrophasors [
7]. They also require a communication backbone capable of transmitting the synchrophasor information at a minimum rate of 120 samples per second, according to IEEE 60255-118-1:2018 [
8]. A 2014 report estimated the cost per unit for Phasor Measurement Units (PMUs), including procurement, installation and commissioning, to be between
$40,000 and
$180,000 [
9]. Due to the high cost of installation and the requirement for a dedicated communication link, the majority of PMU systems have been deployed on transmission networks [
10]. However, there have been attempts to extend the use of PMUs to lower voltage networks. In [
11], incorporating a wireless LAN network to provide low latency communication is proposed. In [
12], a highly optimised, low cost PMU device is developed for the distribution network, but relies on the public internet network for communication which limits its applicability to real-time, low latency applications.
In addition to the lack of synchrophasor measurement capability, it is also often the case that basic real-time monitoring of loading is not available on LV networks. The “DEDUCE” (Determining Electricity Distribution Usage with Consumer Electronics) [
13], which ran as a regulator funded innovation project in in 2017/18, invited University students to submit innovative ideas for low-cost (less than
) LV substation monitors to provide more granular data to the network operator than current solutions. The entries were evaluated against a set of of criteria as part of a competition. The work presented in this paper is based on the initial student-led idea submitted to this competition [
14].
In the UK, a number of other high-profile studies have attempted to tackle the problem of low network visibility at LV level. In [
5], which was a joint study by Western Power Distrubution (WPD) and UKPN, two major Distribution Network Operators (DNOs) in the UK, the current state-of-the-art in LV sensor technology was evaluated. The study’s findings indicated that a satisfactory range of current sensors already exist, but this in itself does not solve the problem of communication, especially from remote LV substations which are out of reach of conventional communication systems like GPRS or ethernet via internet. In [
15], the benefits of improved network visibility on the UKPN network were evaluated. The findings confirmed that there were a range of benefits of improved network visibility, including the deferring of network reinforcement, reductions in customer interruptions and improvements in asset management. However, the granularity of data that can be sent back to the control room is limited by the bandwidth of conventional communication systems, making the implementation of a true real-time monitoring system difficult. In the “Low Voltage Network Templates” project [
16], the real-time voltage profiles across selected parts of WPD’s LV network were recorded. The main conclusion of this work was that a better understanding of the voltage characteristics of the LV network is necessary for DNOs to maintain power quality and cost-effectiveness as more low-carbon technology is added to the grid. In [
17], a decentralised approach to monitoring and communication is proposed. The method works by delegating some of the processing of the raw data to the source of collection, reducing the required bandwidth. However, the cost of additional local processing may provide a barrier to its widespread deployment.
An emerging option for communication of monitoring data is narrowband Power Line Communication (PLC). This technology has recently been standardised and the number of successful deployments of automatic meter reading is growing. The method is a viable means of transmitting data over low voltage networks at bandwidths of less than 1 MHz. However, there is currently limited empirical evidence that the emerging PLC standards (e.g., Prime and G3-PLC [
18,
19,
20]) are able to communicate through transformers and across large MV networks.
This paper proposes a new method which eliminates the need for many of the constituent elements of the traditional monitoring and digital PMU infrastucture, including the communication backbone. Although it is based on a form of PLC, it is analog in nature, which means it is less affected by multipath interference; a particular problem on MV networks [
21,
22,
23]. The method works by frequency modulating a pair of carrier waves with the voltage and current transducer signals. This signal is then amplified and directly re-injected into the network. The power frequency (and harmonics up to a limited frequency) voltage and current waveforms are now encoded as the instantaneous frequency of the injected signal. The receiver demodulates this signal via the Hilbert Transform, which converts the signals back to their original form. Crucially, it is also observed that the phase angle between voltage and current is preserved, providing a low-cost and robust set of synchrophasor measurements across the feeder.
The remainder of the paper begins by describing the proposed method, starting from the basics of FM theory. A simulation model, incorporating EMTP and Matlab, is subsequently developed and used to test the method under realistic conditions. Finally, prototype designs; based on consumer electronics for the transmitter and an FPGA for the receiver, are built and demonstrated under laboratory conditions.
2. Overview of the Proposed Method
The governing equation for FM is:
where
is the modulating signal,
is the frequency of the carrier and
the frequency deviation constant, which may otherwise be expressed as:
where
is the sensitivity of the modulator and
is the amplitude of the modulating signal,
. In the proposed method,
is a replica of the power frequency signal as measured by a transducer and
is the centre frequency of a particular Voltage Controlled Oscillator (VCO) in a range determined by
. This signal can be immediately amplified and re-injected into the power line to propagate to a receiver.
The method of communication is essentially a form of analog FM Power Line Communication (FM-PLC). In contrast to digital PLC, which typically encodes digital data in the phase of a symbol, FM-PLC has no phase transitions between symbols and therefore has the clear advantage of not being affected by Intersymbol-Interference (ISI). As an analog form of modulation, it is impossible to obtain an “exact” reading of voltage or current (ignoring quantisation error), though, as will be shown, a reasonable accuracy can still be achieved.
As an example, consider a signal with a centre frequency,
, of 500 Hz and the signal varies to ±(
) Hz, where
is the modulating signal, in this case the 50/60 Hz power frequency (and harmonics). This can be thought of as a type of frequency modulation because the amplitude of
is encoded in the instantaneous frequency of the modulated waveform. More signals can be encoded by changing the centre frequency such that no two signals overlap. In this case, assigning an
of 1500 Hz to another signal, and retaining the same sensitivity, would not lead to interference, though this assumes a bandlimited power frequency signal. A device capable of outputting a signal whose frequency is proportionate to the voltage of its input is a Voltage Controlled Oscillator (VCO). Such devices can be constructed and easily customised with readily available and inexpensive operational amplifiers and passive components, or off-the-shelf Integrated Circuits (ICs).
Figure 1 shows a high level block diagram of the proposed transmitter architecture. The transmitter supports two channels so can convey both voltage and current information from a single observation point. Because the voltage and current are being modulated simultaneously, the transmitter is also capable of conveying the power factor angle.
It is assumed that there are many transmitters on a network, each occupying a set bandwidth. The high frequency signals from all transmitters are received by a single receiver. This work proposes communication from several monitoring positions on LV networks to a single observation point, most conveniently connected at the primary substation for the feeder. To keep costs to a minimum, we propose the use of the capacitive divider of the Voltage Detecting System (VDS), which are already installed in all major MV switchboards according to IEC 61243-5, for the receiver coupler. This method was recently presented in [
24], with promising results in the kHz bandwidths. Using this method opens up the possibility of receiving the signal on the high voltage side of the transformer, avoiding the extra attenuation and frequency selectivity of the transformer.
Following coupling, the received signal is sent to the receiver.
Figure 2 shows a high level block diagram of the proposed receiver architecture. The first stage in the receiver is a bank of bandpass filters with high pass and low pass cut offs set to match the upper and lower frequencies of each of the transmitters.
To convert the frequency modulated high frequency signal back into the original power frequency signal that is modulating it, the Hilbert Transform (HT) is applied. The HT of a function
is defined as:
where
P represents the Cauchy principal value. Performing the HT on the signal allows it to be treated in the time domain as a rotating vector with instantaneous phase,
and instantaneous amplitude,
. In the time domain, this is given by:
The instantaneous phase is therefore given by:
Equation (
5) can be realised in hardware efficiently using the CORDIC algorithm. Finally, the instantaneous frequency,
, is the rate of change of phase with respect to time:
should be a replica of the original power frequency modulating the VCO transmitter. Further robustness has been achieved by using a final Fast Fourier Transform (FFT) block to determine the magnitude and phase of the power frequency and specified harmonics. If the receiver accepts signals from several transmitters simultaneously, the relative phase differences between the voltage and currents sharing the same transmitter location may be determined. Furthermore, the relative phase differences between transmitters at different locations may also be determined, providing the means to implement a real-time synchrophasor monitoring system without the requirement for time sychnronisation at the transmitters or any other form of communication.
The bandwidth occupied by each transmitter in the proposed scheme is of prime importance because it determines how many devices can operate simultaneously on a given network within a specified total bandwidth allocation. If it is assumed that the modulating signal,
, is given by:
Then the instantaneous phase deviation is given by:
where the modulation index,
is given by:
The FM modulated signal is therefore:
can alternatively be expressed as:
where
is the Bessel function of the first kind. Here it can be seen that the spectrum of
is that of an infinite sum of sinusoids with frequency increasing in multiples of
, multiplied by the Bessel function. It is observed that the value
dictates the strength and spread of the spectra around the centre frequency. Carson’s rule states that approximately 98% of the power in an FM signal lies within
.
At the receiver, all that is known about the incoming signals is (1) The centre frequencies,
and (2) The sensitivities,
. Bandpass filtering isolates each of the
n channels and the HT is performed separately and simultaneously on each. The result of the HT is the analytic signal, which may subsequently be converted to instantaneous phase; a sawtooth like waveform. The differential of the instantaneous phase yields the instantaneous frequency. If the offset is removed such that the waveform oscillates around 0 Hz rather than
for that particular channel and scaled by the sensitivity factor, a replica of the power frequency waveform emerges.
Figure 3 shows a representative set of signals at the receiver, here with an
= 23 kHz.
FM has better resilience to noise and multipath interference than Amplitude Modulation. Such resilience may make FM a good candidate for applications on the power line channel, which are notorious for multipath interference and strong attenuation at particular regions of the spectrum [
21]. In the past decade, narrowband PLC standards based on OFDM communication schemes have emerged. However, such systems require a local MODEM and may still be vulnerable to multipath effects in channels with large RMS delay spreads.
The proposed architectures for the transmitters and receivers have been tested using both Matlab-based simulation models and actual prototypes. For the latter, the transmitter has been realised using a commercially available VCO IC and readily available passive components. The receiver, which is more expensive to realise in hardware, utilises an FPGA in order to implement the necessary digital bandpass filters, HT, FFT and CORDIC algorithms.
3. Development of Simulation Scheme to Test Performance on a Typical MV/LV Feeder
As with any form of communication, the properties of the channel will play a crucial role in the performance of a given scheme. In this case, the power line itself is the channel. There are many studies in the literature which have attempted to characterise the power line channel from a communication perspective, with a high degree of frequency selectivity often cited as a particular challenge. To model the channel accurately, a similar methodology to that proposed in [
21] is used. However, simulation of the proposed method also requires power frequency (and low harmonic) information, posing a difficulty since high frequency models for transformers, loads etc, are not appropriate for power frequency studies. Therefore, we propose the use of two parallel networks which run simultaneously in the same simulation—one for calculating the power frequency information and one for the high frequency communication study. Information from the power frequency study is recorded and passed to the transmitter model, which is implemented in the EMTP Models language. The VCO output of the transmitter model is coupled into the network. The whole process happens timestep by timestep in the simulation.
The simulation methodology used in this work is summarised in
Figure 4. The Matlab domain is responsible for modelling the receiver and calculating the magnitude and phase errors based on comparisons between the actual voltage/currents from EMTP-ATP and the calculated values after propagation through the network and the receiver. The parallel networks within the EMTP-ATP domain is initiated from a DOS command within Matlab. Functions within Matlab have also been written to edit the VCO parameters of the transmitter models, such as centre frequency (
) and deviation (
). Post-simulation, Matlab automatically converts the simulation results from .pl4 format to .MAT format, allowing automated processing.
The simulation network modelled within EMTP-ATP is shown in
Figure 5. It consists of a single primary substation. This has also been labelled as the observation point (“obs”) because the high frequency signals are taken from this position to be sent to the Matlab based receiver. The four three-phase LV feeders are modelled as underground cables (see
Appendix A). The HV feeder runs at a nominal voltage of 10kV and is modelled as a three phase wood pole overhead line (also shown in
Appendix A). In the power frequency simulation, loads (shown as
to
) are represented as RLC elements.
3.1. EMTP-ATP Models
Within EMTP-ATP, two parallel networks are modelled and run simultaneously; one for the power frequency simulation and one for the high frequency simulation. Information from the power frequency solution is passed, during runtime, to VCO transmitter models, which are implemented within EMTP-ATP’s native models language (the code for the Model is shown in
Appendix B). For example, considering the example network, the measured voltage and current waveforms from
,
,
and
, are passed to a series of VCO transmitter models. The output of the VCO models are immediately coupled back into the high frequency network at the same points.
In the high frequency models, extra care must be taken to ensure a representative channel model in the frequencies of interest. For the line and cable models, the JMarti frequency dependent line model is used [
25]. For the transformers, Cataliotti’s medium frequency transformer model is used, which has been shown through empirical testing to perform well up to 100 kHz [
26]. This has been implemented directly in the EMTP using lumped elements.
3.2. Implementation of the Receiver in Matlab
The receiver in the simulation studies is implemented within a Matlab script. To automate the process of performing the EMTP-ATP simulations, the same script is also capable of editing the ATP file (the text description of the circuit to be simulated) and running EMTP using a DOS command. Below is a brief description of each of the main functions within the Matlab script.
Edit_ATP(): To automate and expedite the process of re-configuring the network between successive runs (i.e., to change the load conditions, Transmitter bandwidths, etc.), the ATP file, which contains a description of the circuit according to the EMTP rule book, is automatically re-written from within Matlab. In the code below, the edit function accepts two vectors; and f. is the vector of scaling factors (one for each VCO transmitter in the EMTP network). f is the vector of centre frequencies for each VCO transmitter. A DOS command run within Matlab is used to copy a fresh version of the .atp file for each run of the simulation. This is followed by the edit function, which contains code to edit the atp file by printing the and f values at the appropriate position.
Add_Noise(): This is an optional function which adds a specified amount of AWGN noise to the received signal, enabling Monte Carlo type studies.
Bandpass_Filters(): A bank of configurable bandpass filters were used to split the incoming signal by frequency to isolate the signal sent by each of the n transmitters.
FFT(): Following the Hilbert Transform, each of the n signals should resemble the original power frequency signal at the transmitter. A final FFT is performed on each of the n incoming signals to conveniently calculate the amplitude and phase of the power frequency and, optionally, the harmonics.
Calculate_Error(): An error function is used to compare the receive values with the actual values.
6. Experimental Results
Figure 20 and
Figure 21 shows a block schematic of the experimental setup used to determine the performance of the proposed method. Initially, a single transmitter (single channel) is considered, meaning only magnitude can be assessed at this time. The test setup is comprised of an AC power supply capable of outputting 12 A at 240 V and a variable resistive load bank to produce a variable current at mains frequencies (50 Hz in the United Kingdom). A rogowski coil (as described in
Section 5.1) is used as the transducer. To increase the current, the current carrying conductor is wrapped around the rogowski coil 10 times, increasing the effective range of current to 0→120 A. The output of the rogowski coil is sent into both the Keysight PA2201 Power Analyser (for measurement) and the transmitter, for frequency modulation and subsequent transmission.
Figure 22 shows typical waveforms obtained from signaltap during the course of the experimental testing. Note that a zoomed in view is displayed in the top two graphs in order to preserve detail, but a full cycle of 50 Hz is displayed below (“converted signal”). A notable feature of these results is the high degree of noise in the converted signal (bottom plot). This high level of noise is due the method of differentiation used in the FPGA (simple numerical differentiation). However, the accuracy of the method is dependent more on the fundamental 50 Hz frequency, which becomes apparent after performing the power analysis FFT. Future work will examine more effective methods of differentiation.
Prior to the experimental testing, the receiver was injected with an FM signal generated by a Rigol DG1022 signal generator (which is used as a comparison/benchmark against which the real transmitter can be tested). For this test, the magnitude and frequency of the modulated signal was set to 830 mV and 13 kHz respectively, with the modulating signal derived from the rogowski coil, as specified in
Figure 20.
Figure 23 displays the results for frequency deviations of 4.5 kHz, 3.5 kHz and 2.5 kHz. In all three situations, the relationship between the measured current and the output from the receiver follows a linear relationship, though the slope is different because the effective sensitivity,
is different in each case.
Figure 24 shows the test results for the two experimental setups of
Figure 20, i.e., one case which uses an inductive coupler to link the transmitter and receiver, and the second case which uses a 100 m coaxial cable transmission line. It is observed that both cases result in the same output, indicating that the 100 m line has had no observable adverse effect on the signal. This matches the observation from the simulations, i.e., the channel itself has little effect on the quality of communication.