2.1. Characteristics of Radiometric PD

As previously stated, the charge displaced in an insulation fault results in an electromagnetic signal propagated away from the discharge source. Whilst the actual current discharged in the fault is a pulse with a fall time in the order of 1 to 1000 ns, the resulting radiometric signal bears similarities to a classical decaying oscillation ^[25], with a frequency range in the order of 50–3000 MHz ^[26]. The frequency of the radiometric signal is dependent on the resonant structure of the insulation defect, resulting in a narrow-band band-pass response ^[27]. The propagation environment effectively low-pass filters this frequency range to 50–800 MHz, with the majority of the frequency content residing below 300 MHz.

A variety of radio receivers are capable of detecting and measuring the transmitted PD signal; however, a difficulty of radiometric PD monitoring is susceptibility to any electromagnetic interference. This includes locally transmitted signals within the VHF and UHF bands, such as FM, private mobile radio, digital audio broadcasting (DAB) and digital TV. Figure 4 shows the typical spectrum of broadcast frequencies within the VHF and UHF bands.

Figure 4. VHF/UHF electromagnetic spectrum with typical wide range transmission signals.

FM occupies the 87.7–107.9 MHz band, whilst DAB is typically broadcast within 174–239.2 MHz. Digital TV is broadcast at from 470 to 606 MHz and 614 to 790 MHz. 2G GSM is broadcast at 880–915 MHz and 925–960 MHz; however, this is outside of the band of interest. There are also sources of interference transmitted locally, such as fluorescent lamp ignition ^[28], unlicensed portable radios, such as PMR446, and licensed portable and amateur radio. Fluorescent lamp ignitions should only be a possible issue in indoor environments, in which additional filtering may have to be employed to remove the interference if it is problematic.

Amateur radio, which is transmitted within bands of 50–52 MHz, 70–70.5 MHz, 144–146 MHz, and 430–440 MHz ^[29], requires a license for broadcasting. Radios are also restricted to a maximum transmission power of 10, 17, and 26 dBW for foundation, intermediate and full license types ^[30], corresponding to theoretical distances of 10, 22, and 63 km, respectively, for a receiver sensitivity down to −40 dBm. However, in reality, these distances would be reduced due to the complexity of the propagation path, and it is unlikely that most radio amateurs would transmit at the maximum power allowed. The radiation pattern of the antenna used will also decrease the likelihood of amateur transmission, causing interference to radiometric PD detection.

License-free PMR446 radio is broadcast in a narrow-band of 446.0–446.2 MHz, and has a maximum transmission power limit of −3 dBW ^[31], corresponding to a distance of approximately 2 km. Whilst it is unlikely that these transmissions outside of a radiometric PD monitoring area would be received, there is a possibility that personnel within, or visiting, a HV site may operate personal radios operating in this band. Such transmissions would not be a significant issue, since these types of transmissions would only be temporary, and could be easily discriminated from a source of PD. Interference of this type from outside of the monitoring area would also be easily discriminated when localization was applied, positioning it outside of the HV area.

Ensuring that these locally transmitted signals do not interfere with the radiometric PD measurement requires some knowledge of the typical frequency content of various types of radiometric PD signals, such as those that propagate from insulation faults within transformers, switchgear, and transmission lines. Measured spectra for a variety of simulated defects have been presented in publications for purposes including propagation effects on UHF PD, antenna comparisons for reception, and similarities between galvanic and radiometric PD signals.

The data presented in ^[32] measure the frequency response of radiometric PD within gas-insulated switchgear (GIS) using sulphur-hexaflouride (SF6). The measurements were made using a biconical antenna and a log-periodic antenna, with bandwidths of 30–300 MHz and 80–1000 MHz respectively, at a distance of 2 m from the source of PD. The reported spectra were between 30 and 820 MHz, with the majority of the energy from 30 to 75 MHz. In ^[33], radiometric PD spectra were measured for an air-blast circuit breaker, with a focus on the change in spectra due to the propagation environment. Three sensor positions were used, two of which were fixed at a distance of 12 m from the PD source, and one which was varied between 12 and 18 m. The mean spectra measured at each position fluctuated by over 50 MHz between the three sensor positions; however, the majority of the frequency content was still between 50 and 400 MHz at each position.

In ^[34] and ^[35], a PD fault was emulated using a sandwich of transformer paper submerged in transformer oil. Five sheets were used in ^[34] in between two electrodes paper, whilst eleven sheets were used in ^[35], with the five inner most sheets pierced with a 1 mm hole to produce an air-filled void. The radiometric PD signal produced was then measured using four different antennas, a 5 and 10 cm monopole, a log-periodic and a zig-zag antenna. The cumulative power was measured with and without the PD source in bands of 0–300 MHz and 1300 to 1900 MHz, with the ratio of the two cumulative quantities provided, therefore allowing for the difference in received power to be determined between background broadcast interference and PD within these bands, along with the power level in each band.

The results reported in ^[34] show that the majority of the received power is within the 0–300 MHz band, with the exception of the 5 cm monopole antenna, where the majority of received power was in the 1300–1900 MHz band. However, the reason for this is that the lower bandwidth of such an antenna is restricted due to its physical size, since a monopole antenna should be half the wavelength of the received signal frequency ^[36].

The results shown in ^[35] divide the measured spectrum into 250 MHz bands for each antenna, from DC to 2.5 GHz, with the log-periodic taken as the absolute measurement due to its flat frequency response across the specified range. The antennas were placed at a distance of 45 cm from the PD source, with the exception of the log-periodic antenna, which was placed at a distance of 90 cm to ensure the antenna was in the far field. Approximately 75% of the received power was measured in the 0–250 MHz band, along with 17% in the 250–500 MHz band. The remaining energy was between 1.4% and 0.26% in each consecutive frequency band.

The spectra of three simulated PD faults within a HV transformer were measured in ^[37]—an internal void PD fault, a surface PD fault and a insulator bushing PD fault. The internal PD fault was simulated with a dielectric oil-filled glass vessel placed inside a transformer tank which contained eleven layers of insulation, with a 1 mm hole perforated through the three inner most layers to produce an air-filled void. Two electrodes are connected to each side of the vessel to attach a HV source, the air was removed using a vacuum in order to ensure a lower permittivity withing the void than in the insulating layers.

The surface PD fault was created using a twisted-pair of HV, resin polyamide-imide insulation, enameled wire placed within the tank, which has an additional layer of polyester tris hydroxyethyl isocyanurate over the insulation. Finally, the bushing fault was generated using a porcelain bushing insulator whose surface was covered with a saline solution, which was allowed to dry in order to obtain a salt polluted surface to generate PD. The emulated PD faults were measured using two 10 cm monopole antennas, one placed inside the tank and one outside at a distance of 30 cm from the fault under test. The spectra were measured within three bands—a lower band of 0–300 MHz, a middle band of 300–1200 MHz, and an upper band of 1200–2500 MHz.

The cumulative power within each band was measured, with and without the PD fault, to determine the contribution of locally transmitted interfering transmissions. For the internal PD fault, the percentages of power measured by the inner antenna were 0.3%, 94.3%, and 5.4% for the lower, middle, and upper bands, respectively, whilst the percentages were 61.9%, 35.8%, and 2.1% in the lower, middle, and upper bands for the outer antenna, respectively. For the surface PD fault, the measured powers in each band were 0.1%, 99.1%, and 0.8%, and 89.2%, 10.7%, and 0.1%, in the lower, middle, and upper bands for the inner and outer antennas, respectively.

The cumulative energies for the bushing PD fault in the lower, middle, and upper bands were 69.3%, 27.4%, and 3.3% for the inner antenna, and 40.3%, 58.1%, and 1.6% for the outer antenna. For the two internal fault types, the majority of the cumulative power measured by the internal antenna was in the middle band, whilst for the external antenna, the majority of power was in the lower band. The inverse of this is the case for the bushing fault, where the majority of the power was in the middle band. The reason for this was due to the transformer tank causing low-pass filtering to the radiometric signal. However, the higher frequency content of the measurements made where the tank was not an obstruction was closely related. A measurement made at a distance greater than 3 m would likely have incurred attenuation to the higher frequency content.

In ^[38], the spectra of an emulated HV floating electrode PD fault, which is a common fault in switchgear ^[39], was obtained and compared using both galvanic and radiometric measurements.

A floating electrode PD test cell was energized to 6.2 kV using a HV DC supply. The measured energy was divided into bands of 50–290, 290–470, and 470–800 MHz. For the contact and radiometric measurements, the percentages of energy in each band were 62.8% and 78.3% in the 50–290 MHz band, 0.76% and 4.33% in the 290–470 MHz band, and 0.56% and 1.59% in the 470–800 MHz band, respectively, and the remaining energy was below 50 MHz. In both cases, the majority of the measured energy was within the 50–290 MHz band.

In ^[20][40], spectra were taken for a further two emulated PD faults using galvanic and radiometric measurements. Both PD cells emulated defects within solid insulation. One consisted of an acrylic tube containing two electrodes at each end, between the electrodes a sandwich of three circular pieces of 1.5 mm thick FR4 insulation are placed, in which the middle piece has a 1 mm hole drilled in the center to create a higher electric field within it, and therefore, a discharge. The second consisted of a sandwich of three sheets of 2.4 mm-thick epoxy glass dielectric between two electrodes, with a 1 mm diameter hole, again, drilled in the center of the middle sheet.

The acrylic tube PD source was filled with transformer oil in order to suppress any discharges occurring at the edges of the electrodes. However, measurements were taken with and without the presence of oil. The acrylic tube emulator was energized to 20 kV AC, both with and without oil, to generate PD, whilst the epoxy glass dielectric emulator was energized to 18 kV AC. Even though the frequency response of the acrylic tube emulator was higher with the presence of oil, almost the entire energy is in the 50–800 MHz band for all three measurements of the two emulators, with the majority of the energy being below 300 MHz.

2.2. Time Difference of Arrival PD Measurement

TOA and TDOA techniques, traditionally used for radio positioning systems, use the times that signals are received at specific receivers to determine the distance, using the speed of light to convert time to distance, of a transmitting signal from each receiver, and therefore, the location of the transmitting source ^[41]. TOA relies on the time at which the transmitted signal was sent along with the reception time for each receiver, whereas TDOA only requires the latter, and is therefore more versatile ^[42]. Figure 5 shows an example of TDOA positioning.

Figure 5. Time difference of arrival (TDOA) location of a wireless transmission with multilateration ^[42].

In this example, four receivers are positioned around a transmitting signal which is the target for location. The distance of the source from each receiver is calculated by determining the precise time difference of received signals between sensors. This is then converted to distance using the speed of light. The resulting distances are used to plot hyperbolic lines via non-linear regression, at which the intersecting point of the lines is the estimation for the transmission ^[43]. TDOA has been used in many schemes for PD location and can provide accuracy within 5 cm ^[44].

TDOA techniques have been employed for the detection of PD using multiple techniques, such as cross-correlation, cumulative energy, and the amplitude of the first received peak ^[45][46], that is, by comparing received PD patterns, or by using the integral of the received signal or the amplitude of the first received peak. TDOA location provides a high level of estimation accuracy and has been used to successfully detect and locate sources of PD in live HV environments ^{[47][48][49][50][51]}.

Although TDOA techniques provide an accurate and non-invasive solution to PD detection and location, there are various constraints that impact the feasibility of large-scale deployment using this technology. The sample-rate required to discern the differences has to be as high as possible, since the time between samples equates to the resolution in distance ^[52]. Since the transmitted signal is a radiometric electromagnetic wave, the resolution in distance is proportional to the speed of light divided by the sample rate. For example, a sample rate of 1 GSa/s has a resolution distance of 30 cm, whereas a sample rate of 5 GSa/s has a distance resolution of 6 cm ^[53]. Whilst techniques such as interpolation can be used to increase the distance resolution at reduced sample rates ^[54][55], rates in excess of 1 GSa/s are still necessary.

The requirement for high sample rates makes this technology less attractive for a flexible large-scale battery powered wireless sensor network (WSN), since the power requirements of such a high-speed data processing system would not be capable of sustained operation from a single battery source over a reasonable period of time and would also have considerable costs. Furthermore, the complexity of scaling a coherent TDOA system on a large scale would be difficult, due to requirement for synchronization between nodes to accurately determine the time difference of signal reception.