Light power resource availability for energy harvesting photovoltaics for self-powered IoT

As the Internet of Things (IoT) expands, the need for energy-efficient, self-powered devices increases and so a better understanding of the available energy resource is necessary. We examine the light power resource availability for energy harvesting photovoltaics (PV) in various environments and its potential for self-powered IoT applications. We analyse light sources, considering spectral distribution, intensity, and temporal variations, and evaluate the impact of location, seasonal variation, and time of day on light power availability. Additionally, we discuss human and building design factors, such as occupancy, room aspect, sensor placement, and décor, which influence light energy availability and therefore power for IoT electronics. We propose a best-case and non-ideal scenario in terms of light resource for energy-harvesting, and using a commercially available organic PV cell, show that the energy yield generated and available to the IoT electronics, can be anywhere between 0.7 mWh and 75 mWh per day, depending on the lighting conditions.


Introduction
Billions of Internet of Things (IoT) devices are predicted to be installed within the coming decade and nearly half of those devices will be deployed indoors [1].The improvement in functionality of these devices has resulted in a decrease in overall power consumption and has coincided with the development of several photovoltaic (PV) technologies, such as Dye sensitized solar cells [2], organic solar cells [3], and perovskite solar cells [4], that are highly efficient at harvesting ambient light, resulting in the possibility of developing perpetually self-powered battery-less IoT.One of the advantages these newer generation of PVs have over more established technologies is the ability to tune the spectral responsiveness of the PV device tailored to varied light conditions [5] found in homes [6][7][8] offices [9], factories [10,11], hospitals [12][13][14], retail stores [15,16], and other indoor locations [17][18][19][20][21][22][23][24].
The types of sensors employed in IoT vary widely in both their function and power consumption.These include temperature sensors (lowest power consumption reported at 71 nW [25]), motion sensors (32 µW [26]), light sensors (57.6 µW [27]), gas sensors (1-100 mW [28]), and humidity sensors (150 µW [29]).Different communication protocols also have varied power consumption which we have summarized in previous work [30].Ultimately the overall power consumption will be a balance between what is required (i.e.how often to 'sense' and transmit data) and how much power is available in certain location.Algorithmic control based on the available light resource will almost certainly be a feature of self-powered IoT.
Characterizing these Indoor PVs (IPVs) is a major challenge as this is typically carried out under low illuminance artificial light [31].The recognized standard for solar cell testing for outdoor applications is ISO 9845-1:2022 and at the time when the majority of this work was carried out, no such standard existed for ambient light testing.Fortunately, in the intervening period between submission, review and final publication a standard on indoor PVs testing has been published, IEC TS 62607-7-2:2023, allowing us to comment on our findings and how this relates to the newly published standard.
One reason that the standard took a long time to materialize is that unlike the sun, the variable spectral characteristics of ambient indoor light meant it was difficult to reach a consensus on what exactly is meant by the term 'ambient light' .For example, office lighting often makes use of diffuse light to provide safe and comfortable working environments [32] or utilizes blue-enriched lighting to improve alertness [33].Hospital lighting may incorporate relatively high levels of UV-A (315 nm to 400 nm) to disinfect surfaces and destroy pathogens [12,34] or may optimize lighting to assist clinicians in promoting healthy circadian rhythms for patient recovery [35].Supermarkets also modify the colour of lighting to prolong shelf-life and influence our perception of fresh produce [36][37][38][39].
The lack of agreement on the spectral characteristics of measurement standard was further complicated by the common usage of illuminance values (lux) to describe ambient light intensities instead of absolute spectral irradiance (usually mW cm −2 nm −1 ).Illuminance is a measurement of intensity, as perceived by the human eye typically between wavelength of 400 nm-700 nm, whereas spectral irradiance is a measurement of power density across a wider range of wavelengths (typically, 300 nm-1500 nm).Since many solar cells technologies absorb light above 700 nm, claims of record-breaking ambient light solar cell efficiencies must be treated with caution.
A variety of illuminance values are commonly chosen by researchers (figure S1 [3,).The two most common are 200 lux and 1000 lux and it is not entirely clear why these values have been chosen.Perhaps researchers have been influenced by the standard, EN 12464-1 [74], which recommends that 200 lux is specified as the minimum lighting standard in public venues such as lecture theatres, public lounges and transportation hubs, while 1000 lux seems to be a value specified for specific inspection tasks.If this is the origin of such values, then it must be noted that EN 12464-1 refers to the illuminance level perceived by the human operator at a workstation performing a specific task and this is not necessarily where you might expect to find or place an energy harvesting IoT node.
In order to aid our understanding of the light resource availability in typical locations a comprehensive study of different lighting environments in offices and laboratories located on Swansea University's Bay campus was undertaken.These buildings were constructed between 2015 and 2019 and, as such, should conform to the regulatory lighting standards set out in EN 12464-1 and so the locations chosen in this work can be considered as typical for modern office environments.
Logging illuminance meters are placed in locations where one would expect to find an IoT node, i.e. on walls and ceilings, and the illuminance values logged over periods of several days.This work will show that in these locations, illuminance values are often lower than that specified by the standards meaning that researchers developing PV cells for ambient light energy harvesting could be overestimating the light resource availability when giving typical performance outputs.The effects on light resource availability due to human factors, such as office occupancy, location of IoT node and interior decoration are also explored and discussed.

Experimental
The low light solar simulator consists of a Thorlabs DC2200-High-Power 1-Channel LED Driver with Pulse Modulation to drive an LED array fitted with an acrylic diffuser.A Keithley 236 source meter is used to measure IV curves.Illuminance values are logged using Onset HOBO MX2202 illuminace meters.For in-situ room illuminance measurements, the MX2202 illuminace meters are configured to log illuminances at 1-minute intervals.They are deployed on various indoor surfaces-i.e.walls, ceilings and desks-for several days.Deployment locations are chosen to be representative of potential IoT node positions (e.g.sensors, communication beacons).To determine the correction factor for the illuminace meters, their response is measured under a low-light conditions in a solar simulator and compared to the response of a CEM DT8809A handheld illuminance meter.A calibration factor of 1.5 was determined with an introduced error of 2% for quoted lux values (figure S3).For experiments to determine the directionality of ambient light, a GY30 (BH1750FVI) digital ambient light intensity sensor mounted on an SG90 servomotor is used.The servomotor is rotated in increments of 1 degree, taking the GY30 through a full 180 degrees on the vertical plane.Spectral irradiance is measured using a factory calibrated Ocean Insight FLAME VIS-NIR spectrometer with diffuser (diameter 7140 µm).The spectral irradiance of dimmable ambient LED room light is measured using the Ocean Insight FLAME VIR-NIR spectrometer.The spectral irradiance of other common indoor light sources-namely CCFL, halogen lamp and natural daylight-are measured for comparison.These irradiance spectra are then integrated between 400 nm and 820 nm to give their respective absolute irradiances.Room décor is modelled within the low light solar simulator by placing coloured card on three walls of the solar simulator.A MX2202 illuminance meter and Ocean Insight spectrometer are placed in the centre of the solar simulator on the same illumination plane (approx.15 cm from the diffused LED light source).IPV external quantum efficiency (EQE) measurements are carried out using a custom built EQE measurement system in AC mode.Prior to performing the measurements, the spectrum of the lamp is measured using a calibrated silicon reference cell.The IPV organic photovoltic modules used for energy yield calculations are supplied by Epishine (part number: LEH350X50610).They consist of six cells and cover an area of 50 mm × 50 mm, typically providing an open circuit voltage of 3.8 V with short circuit current of 147 µA at 500 lux warm white LED on a white background.Amorphous silicon solar cells (AM-1417) are supplied by Sanyo, providing an open circuit voltage of 2.4 V and short circuit current of 13.5 µA at 200 lux fluorescent white light.

Results & discussion
PV efficiency is a simple ratio of the electrical power density output of a PV device and the irradiance power density of the incident light.Indoor and ambient light PVs, however, are almost exclusively measured using illuminance values (lux).Illuminance, based on the spectral response of the human-eye, is considered a poor way to estimate irradiance; however, illuminance meters are relatively cheap and widely available and so are often employed to characterize ambient light sources.Ambient light sources can be natural sunlight filtered by glazing, LED lighting and increasingly less common, fluorescent and halogen lighting, all shown in figure 1 below.Figure 1 shows the relationship between the illuminance values of a range light sources compared to their spectral irradiances and it can be observed that for these particular light sources, over the range of illuminances measured, the spectral distribution scales linearly with light intensity, i.e. the spectral 'shape' is maintained.Care must still be taken, however, when determining cell performance under ambient conditions with LED lighting as it is known that LED spectral distribution is not always consistent across varying light intensities due to changing driving currents altering the junction temperature [75].The spectral match between the emission light source and the indoor PV cell's light absorption characteristics, determined by its band gap, has been the topic of much discussion in recent years, but the adoption of LED lighting over fluorescent and incandescent light sources has allowed PV researchers to focus their efforts in designing materials for optimum absorption for a mixture of artificial and diffuse natural lighting.
When employing illuminance meters to estimate irradiance, it is advised to proceed with caution as there can be large discrepancies between different illuminance meters under identical test conditions.In this work we used HOBO MX2202 illuminance meters from ONSET.We compared these under test to a handheld illuminance meter (CEM DT8809A) in a low illuminance solar simulator.The data is shown in figure S2 where it can be seen MX2202 datalogger illuminance measurements are observed to be 40% lower than the DT8809A measurements.This is because the MX2202 under-reads illuminance due to the lack of cosine correction.As the response of both devices are linear within the light intensity regime measured, we can apply a correction factor of 1.50 to the MX2202 datalogger measurements for a more accurate estimation of real illuminance values.Another factor to consider is repeatability across different devices of the same type.In figure S3 we show measurements of five identical MX2202 illuminance meters in the solar simulator show a spread of 5% in their measured illuminance values across the light intensities considered.Despite these issues, because of their low-cost and ease of use, several datalogging illuminance meters (MX2202) can be deployed simultaneously across multiple indoor locations whereas the cost of doing so with spectroradiometers would have been prohibitive.These locations chosen include offices, laboratories, corridors, and stairwells.The loggers were positioned in places where one might consider placing wireless IoT nodes; places where they will be most unobtrusive, inconspicuous and discreet.
Figure 2(a) shows a typical open plan office with zoned lighting.The locations of the logging illuminance meters are shown on the photograph and designated as W (office wall), K (office kitchen) and C (office celling).Figures 2(c) and (d) shows the illuminance measured at the locations over two different time periods.The data illustrates several important factors when considering the deployment of indoor PVs in the office environment.Firstly, the maximum illuminance measured is 150 lux and, in some locations, 50 lux, far below recommended task-oriented lighting standard.In this work it has been found that wall illuminance of 50 lux is reasonably common in office-based lighting and so it is therefore recommended that researchers working on indoor PV materials and devices consider measuring their cells efficiencies at 50 lux, or at least investigate the low-light operating limit of their devices as there is a potential that devices that appear to perform well at 200 lux, may not generate any power at 50 lux or below and so would be unsuitable for powering IoT sensor nodes in these very low light locations.
In modern buildings, spaces are lit by a mixture of LED lighting and natural daylight.Spatial distribution of illuminance (and therefore irradiance) varies considerably because of human-behavioural and building design factors.For example, in some laboratories, smart lighting automatically dims according to the  amount of daylight entering the room and switches off when no movement is detected after a long period of time.In larger open plan offices, lights are manually controlled so that sections of the office are selectively illuminated, according to current occupation.This is illustrated in figures 2(c) and (d) where the effects of zoned lighting can be observed, where certain locations experience very low illuminance, dependent on the occupancy of different zones of the office, at different times of day.This effect is particularly highlighted in figure 2(d) where all zones measured are <20 lux when the office is vacated over the weekend and because of this self-powered IoT devices must be designed with energy storage elements to provide enough power over a weekend when such office spaces are vacated.The relationship between light resource availability and human-behaviour is perhaps best shown in figures 2(b) and (e).Illuminance meters were deployed in a corridor environment where the lighting is controlled by passive infrared (PIR) sensor.The light resource availability is highly dependent on human activity and for periods of time the corridor experienced almost complete darkness.An understanding of human behaviour in the environment and the occupancy frequency of certain locations will be critical in predicting available light to power IoT nodes.Other factors that affect light resource availability include the position of the IoT node with respect to light sources in that location.In a typical working environment, lighting can be classified into four categories: 1. Direct artificial light, where the light fittings cast beams onto walls and floors, e.g.spotlights.This kind of lighting is typically found in corridors and toilets and results in non-uniform illumination of surfaces.2. Diffuse artificial light, where light fittings are intentionally diffuse to give more uniform lighting throughout the room.This lighting is most often found in offices and laboratories.3. Natural light, which can be either direct sunlight or diffuse daylight.4. A superposition of two or more of the above.
The greatest variability in illuminance comes from rooms with external windows where seasonal changes dictate the amount of natural light entering the room.Although there is more sunlight in the summer months, the sun is higher in the sky, so there is limited direct sunlight entering south-facing rooms.However, in the winter months, the sun is much lower in the sky, allowing direct sunlight to penetrate farther into south-facing rooms on cloudless days.Other human factors come into play, e.g.occupants closing blinds to prevent screen glare drastically reduces the amount of natural light available for harvesting.Figure 3 shows data from an office with a south facing window.The data was recorded in August at a latitude of 51.6 • and longitude of −3.9 • and so the maximum solar elevation angle was approximately 55 • .This means the sun does not shine directly into the office at any point during the day.The data shows two consecutive days: one sunny and the next overcast.It can be observed that the highest illuminances measured are when the meter is placed horizontally on the desk.This result is to be expected as the illuminance meter is directly incident to the artificial lighting and has a favourable position with respect to the window and the angle of incidence for natural light.The result is also concurrent with indoor lighting standards which are focused primarily on the illuminance experienced by the worker at their workstation.Why then, do we not simply place light harvesting IoT nodes on workstations and desks to take advantage of the higher illuminance values?Apart from the fact that self-powered IoT nodes are likely to be passive sensors or low-power communication relays, requiring them to be unobtrusive and discreet, the answer can be seen in figure 3(d), where the measured illuminance is prone to sudden and prolonged dropouts.Desktops and workstations can be cluttered with objects, e.g.papers, books, coffee mugs, etc, and these could occasionally obscure any desktop mounted IPV devices.This can be seen in figure 3(d), where the cause of the dropout is due to the author accidentally placing their notepad over the illuminance meter.This again shows how critical human behaviour is to light resource availability.While placing a self-powered IoT node on a workstation or desktop may seem like the best option in terms of light resource, it may not be the ideal location in terms of interior design factors and indeed the IPV being covered due to human error.Figure 3(b) shows the data from the same room with desktop illuminance data removed for greater clarity.It can be observed that the aspect and location of the illuminance meters has very little influence over the illuminance values measured.This can be explained by the fact that apart from the desktop meter, none of the remaining meters have either natural or LED lighting directly incident onto their detectors.They are all therefore measuring diffuse light.The fact that the walls of the office are white means that the light resource is scattered and uniformly distributed regardless of location within the office.
While it may be obvious that a room with a south-facing window experiences significantly higher illuminance than a room without any natural light, further consideration must be applied to the different spectral characteristics, and how these might change during the day.In modern workspaces without natural light the spectral illuminance will be dominated by white LED lighting, whilst the workspace with mixed lighting will experience a change in the spectral characteristics throughout the day and with varying meteorological conditions.This is illustrated in figure 4 which shows the spectral evolution of light in a room with a south facing window, throughout the course of a typical day.The room is illuminated by daylight as well as LED lights that automatically brighten and dim according to the amount of daylight that is available within the room, further adding to the spectral complexity.This can be observed in figure 4, when in the early morning, the spectrum is dominated by natural light.As the room becomes occupied later in the morning, the PIR sensors switch on the LED lighting and the LED blue emitter peak becomes identifiable in the mixed spectrum and by late afternoon the spectrum is dominated by LED lighting and the emitter and phosphor peaks are clearly identifiable in the spectrum.Figure 4(c) shows the averaged spectral irradiance throughout the day and while the influence of the natural light spectrum can be observed, it is the spectral distribution of the LED lighting that appears to dominate.The position of the self-powered IoT node within a space also becomes important in such mixed light environments.Figure 4(b) shows spectral variance as a function of distance from the window.The spectra were recorded at 10:30, and as one would expect, the natural daylight spectrum dominates nearest to the window, while the LED spectrum is dominant furthest from the window.
The spectral characteristics of the light source are important when considering the band gap of light-absorber materials to use in energy harvesting PVs.For example, LED lighting with colour temperatures ranging from 1700 K to 6500 K would correspond to ideal band gaps in solar cell materials of 1.60 eV-1.97 eV [76].For natural daylight filtered by modern building glazing, a band gap of 1.10 eV [71] would be more suitable.To calculate the optimum band gap for the scenario outlined in figure 4, a detailed balance analysis is used, based on that originally outlined by Shockley and Queisser [77], the optimal band gap of the PV system for the incident spectra can be calculated and for the spectra shown in figure 4(b), the optimal band gap varies slightly (1.7 eV-1.8 eV, see figure S6) depending on the amount of daylight present in the spectrum.The thermodynamic limit of the power conversion efficiency (PCE) likewise changes from 37% to 39% at an illuminance of 100 Lux.The full details of these calculations can be found in a recent work [78].These upper limits of PCEs represent the maximum achievable values and do not include losses associated with parasitic resistances, defects, or device size [30].In practice, tuning the band gap for each individual environment would be impractical and having a materials system that is easy to manufacture, stable and able to operate sufficiently (even if not optimally) in low-light environments is more important than designing the ideal band gap for each individual scenario.
Lighting in offices and laboratories designed to radiate out at certain angular distributions, which could result in a variance in light resource that is height dependent.Figure 5(a) shows the variance in illuminance as a function of height below the ceiling.In this location ceiling mounted LED luminaires produce diffuse lighting which illuminates the walls.Illuminance can vary by as much as 30% dependent on vertical height placement on the wall, this is something that installers of these technologies should be aware of in optimizing the positioning of self-powered IoT nodes.In most scenarios, to maximize light harvesting opportunity for wall mounted devices, it would be unfavourable to have the device mounted parallel to the wall but instead angled to some degree toward the ceiling mounted lighting.Figures 5(b) and (c) show the angular dependence vs the height below ceiling in two different locations.In figure 5(b), the LED luminaires are 0.6 m from the wall and in figure 5(c) the luminaires are 1.2 m from the wall.It can be observed that maximum illuminance is dependent on both the height below the ceiling and horizontal distance from light fittings.In figure 5(b), maximum illuminance is observed at an angle of 70 • and is practically independent of height.In figure 5(c), maximum illuminance is observed at an angle between 34 • to 54 • , depending on height.It can be concluded therefore, that angular dependence becomes more critical with increasing horizontal distance from the ceiling light source.Given the data in figures 5(b) and (c), it would be recommended that designers of indoor PV powered IoT nodes take this into consideration by choosing a fixed angle of, for example 45 • or designing a tilt system to allow maximum light harvesting in a range of scenarios.
Differences in reflectivity and colour are essential design elements within modern buildings because they provide vital contrast.Colour also influences the subjective visual perception of light, and is associated with psychological, physiological, and social reactions [79].In architecture, Light Reflectance Values are a measure of the percentage of visible and usable light that is reflected from a surface when illuminated by a light source [80].Interior designers use these values to design spaces that meet indoor lighting standards.To investigate whether room décor influences light resource availability, the spectral irradiance was measured in two otherwise identical offices, one with blue walls and one with green walls.Figure 5(d) shows the normalised irradiance and indeed there are subtle differences in the measured light spectrum.The office with green walls shows reduction in the peak associated with the LED emitter and increased apparent irradiance in the region associated with the LED phosphor, indicating an increased absorption (by the walls) in the blue portion spectrum and increased reflection in the green portion of the spectrum.Irradiance measured in the blue office shows no change from the emitter peak but decreased emission in the phosphor region due to increased absorption of green light by the walls.The spectral variations due to room décor may influence predicted PV performance due to spectral mismatches with the PV cell's band gap, as discussed previously.To investigate further, coloured card was used to surround the measurement area of a solar simulator set up to measure PV cells at low light.Before measuring PV cells, a spectrometer and illuminance meter were placed in the solar simulator on the same horizontal plane, facing the LED light source.Light is received by the spectrometer and illuminance meter as a superposition of direct LED light that is generally normal to the receiving devices, while diffuse reflected light from the solar simulator walls reaches the receiving devices at more oblique angles of incidence.Even though most light is received directly from the LED light source, the irradiance measurements, shown in figure 5(e), show clear spectral dependence with red and orange background card resulting in higher apparent irradiances compared to blue and green backgrounds.
The appropriate PV technologies are crucial for energy harvesting strategies in IoT applications.Two types of PV module commonly specified for energy harvesting IoT, organic photovoltaics (OPV) modules and amorphous silicon (a-Si) modules, were then measured in the solar simulator with the coloured card backgrounds and the results are shown in figure 5(f), where the maximum power output (PMAX) of the module is shown as a % loss compared to the module's PMAX measured with a white background.Coloured walls reflect less radiant energy than white walls, so in all cases, PMAX is less than what would have been achievable with white walls.Red and orange walls reflect more light at the red end of the irradiance spectrum and since the OPV module's spectral responsiveness (external quantum efficiency-see figure S4) extends further into this longer wavelength region, it is more efficient at harvesting this energy than the a-Si whose spectral responsiveness which is reduced significantly at wavelengths >550 nm, so is unable to harvest as much of this longer wavelength energy.Given these results the recommendation to interior designers, to maximise light harvesting efficiency for IoT objects, therefore, is to paint your walls white, but if you must add colour, choose colours at the red end of the spectrum!Next-generation PV technologies such as OPV or perovskite PV (PPV) offer several advantages for their application to IoT.Firstly, as solution-processable technologies, they are amenable to low-energy manufacturing techniques.This provides opportunities for lowering the overall energy footprint of IoT devices.Secondly, the band gap tuneability of next-generation PV materials such as OPV and PPV means they can be optimized for various lighting scenarios.Under typical indoor light sources, IPVs have the potential to achieve PCEs surpassing 50%, which is significantly larger than the predicted PCE of 33.7% for AM1.5G sunlight.However, to achieve such high PCEs, semiconductors with wider optical band gaps between 1.7 eV and 1.9 eV are required.This is considerably wider than traditional solar cell materials such as crystalline silicon, gallium arsenide, and cadmium telluride [81].
As previously mentioned, during the submission and review of this manuscript, the indoor PV measurement standard (IEC TS 62607-7-2:2023) [82] was released by the International Electrotechnical Commission and we would like to briefly comment on the standard with respect to our findings.The standard recommends two light sources, one fluorescent and one LED, and measurements at 1000 lux, 200 lux and 50 lux.We are pleased to see the inclusion of the 50 lux recommendation as our data shows this is not an uncommon illuminance value for an IoT node that might be placed on a wall, and we have already referred to why good IPV performance at 50 lux may be important.Additionally, it is not a value that many in the IPV community readily choose for their cell measurements, and we are happy that the authors of the standard recommend that they should.The choice of including fluorescent lighting is interesting as whilst still very common, fluorescent lighting is mostly being phased out in favour of LEDs, for example, the Swansea University campus we used for this work was built in 2015 and we could not find a single installation of fluorescent lighting in any of the locations considered across the campus estate.There is also no mention of the mixed spectrum that would be a feature of any room with a window, but this is understandable given the complexity of the issue and when one considers the main purpose of the standard, at least from the perspective of the IPV community, is to be able to reliably compare measurements across different laboratories, and so the spectral characteristics should be specified for a single light source just as they are for AM1.5.In a similar fashion the standard recommends collimated as opposed to diffuse light, and whilst a diffuse light source might be more realistic in terms of the indoor PV scenario, a reliable, repeatable diffuse measurement may be be more difficult to recreate across multiple laboratories than a collimated measurement.Lastly the LED light source chosen is CIE LED-B4 which has a colour temperature of approximately 5000 K, a colour temperature typically called 'cool white' .A slightly warmer colour temperature (4000 K) is more common in office spaces, warmer still (2700 K) for domestic environments.
However, 5000 K is common in hospitals and other clinical settings and ultimately a standard must be chosen for reliable comparison within the community.After all, we still use AM1.5 to characterise our outdoor solar cells, even though in Swansea, sitting at 52 • on Europe's North Atlantic coast, we may only experience '1-Sun' for a few minutes each day in June-if we are lucky!So far, this work has focussed on the variability of the light resource available to power energy harvesting PVs, designed to power the next generation of self-powered IoT nodes.What does this mean therefore for the designers of low-power electronics that will be used in conjunction with energy harvesting solar cells?How much usable electrical power is available considering the light resource availability, the light harvesting efficiency of the PV module and the efficiency of the energy harvesting electronics?To answer this question, it is perhaps best to identify a best-case but realistic scenario, and a non-ideal scenario.In the best-case scenario, we have identified a semiconductor manufacturing clean room-see figure S5.In this scenario, no natural light is present and so higher light intensities may be available in locations with a large degree of natural light.However, in the clean room, the lighting is relatively bright for artificial light and on 24 h per day and so is very predictable and stable.A wall mounted IoT node can expect to receive illuminance of around 3300 lux, 24 h per day, seven days a week.To evaluate the usable electrical power, we used Epishine OPV modules as these are one of the only OPV modules that are commercially available and made specifically for indoor lighting conditions.Under the clean-room test conditions the maximum power output of the Epishine OPV module is measured as 3477 µW.If we assume we have the latest energy harvesting electronics operating at an efficiency of 90%, then 3129 µW of useable power could be available to power the IoT electronics.We will propose that the non-ideal case to be the corridor without natural light, shown in figure 2(b) and (e), where an average 80 lux can be expected for a device on the wall.The Epishine OPV module gives a power output of 91 µW (82 µW available for the electronics), but the light here is also determined by the number of people using the corridor.In our data the lighting is on approximately 50% of the time, Monday to Friday.The corridor appears to have no traffic at the weekends and so illuminance is negligible.Total hours of available light is therefore 60 h.When illumination is not constant, as it is in the cleanroom, it is perhaps more useful to think in terms of total energy yield over a period of time, rather than power output at a given point in time.Therefore, the total useable energy yield for device in the corridor is 4.9 mWh per week vs 526 mWh per week for devices located in the cleanroom.This results in an average daily energy yield of 0.7 mWh per day (∼0.98 mWh d −1 Mon-Fri, 0 mWh d −1 at weekends) and 75 mWh per day respectively.
The PV power output at a specific condition and the average energy yield are important parameters to consider when designing self-powered IoT nodes, in what can be powered and what the energy storage requirements may be.For manufacturers of self-powered IoT nodes, it may be necessary to specify minimum lighting conditions to potential customers, in order to avoid devices consuming more power than can be generated by the solar cell.

Conclusions
We have shown that predicting the performance of energy-harvesting PVs for indoor applications is significantly more complex than for outdoor PV cells.We have also shown that in many cases the realistic light resource can be much lower than many researchers of ambient-light PV may realise and that they must consider measuring the performance of their PV cells down to 50 lux or even less.The primary source of this complexity stems from the variability in light power resource, which is influenced by factors such as spectral distribution, intensity, and temporal and seasonal fluctuations.Moreover, human and architectural elements, including room orientation, IoT node placement relative to windows, presence of PIR sensors, and room occupancy, further compound the intricacy of performance prediction.To accurately estimate the energy yield for PVs powering IoT nodes, all of these factors must be taken into consideration.To estimate the range of energy that can be generated, we proposed best-case and non-ideal scenarios and using a commercially available Epishine OPV cell, show that the energy yield generated and available to the IoT electronics, can be anywhere between 4.9 mWh and 526 mWh per week, depending on the lighting conditions and human and architectural factors.The data presented here will prove valuable for the design of ambient light-harvesting PVs, as the observed spectral variations can guide the selection of an optimal band gap for maximum PV cell efficiency.Additionally, electronic engineers can utilize this information to better comprehend the power availability under both ideal and non-ideal conditions, thus facilitating more informed design and power management decisions for PV-powered ambient light-harvesting IoT nodes.
Reflecting on the variability of light power resources, it becomes evident that approaches to ambient light powered IoT requires a nuanced perspective.Electronic engineers will need to develop adaptive IoT systems that adjust their power consumption in response to harvested energy so they can ensure consistent functionality in diverse light conditions.Delving deeper into human behaviour patterns will shed light on the intricacies of light availability, fostering more intuitive energy harvesting strategies.Additionally, utilising machine learning or AI algorithms to analyse varied environmental data may pinpoint optimal placement of IoT nodes, enhancing energy capture and device efficiency.

Figure 1 .
Figure 1.Spectral irradiance at a range of illuminance values for (a) natural daylight measured through glazing; (b) cool white LED, colour temperature 5000 K; (c) compact fluorescent tube; (d) halogen lighting.

Figure 2 .
Figure 2. Illuminance meter locations in (a) office environment, C = ceiling, W = wall, K = kitchen wall; and (b) corridor with no natural light.(c)-(e) illustrate how energy efficient lighting controlled by PIR sensors and combined with human behaviour can lead to complex and difficult to predict light energy resource availability.

Figure 3 .
Figure 3. (a) and (b) Illuminance meter data from several locations within (c), an office with south facing windows.(d) The effects of human error in accidently covering a self-powered IoT node.

Figure 4 .
Figure 4. (a) Evolution of light spectra during a typical August day in an office with south facing windows.(b) Spectral variation with distance from the window at 10:30.(c) Average lighting spectrum, compared to spectra at 09:00 and 21:00.

Figure 5 .
Figure 5. (a) Illuminance as a function of height below ceiling; (b) and (c) illuminance as a function of PV module angle from the vertical (i.e.parallel to wall), (b) = 0.6 m and (c) = 1.2 m horizontal distance from LED luminaires; (d) normalized irradiance spectra in two identical meeting rooms, with different green and blue coloured walls; (e) irradiance spectra inside a low light solar simulator with different coloured walls; (f) calculated percentage PMAX loss compared to PMAX measured with white walls, of an OPV module and an a-Si module in low-light solar simulator-with coloured walls.