Motivation
Shared utility metering is the norm in Malawian multi-tenant housing ("compound housing"), where a single ESCOM prepaid meter serves between two and ten households. Cost apportionment is typically negotiated — per room, per appliance count, or per head — and the resulting disputes are a persistent source of tenant-landlord friction. Certified submeters exist but their per-point cost (US$120–180 installed) exceeds what the rental economics of compound housing can absorb.
The question this work addresses: how accurate and how durable can a submetering point be at a materials cost an order of magnitude lower, and is that accuracy sufficient to displace heuristic apportionment in practice?
System design
Each metering point comprises a split-core current transformer (SCT-013-030) on the tenant's feed, an ESP32 module sampling at 2 kHz, and a 25 F supercapacitor bank providing write-through persistence across outages. Points report over 433 MHz LoRa to a compound-level hub, which handles prepaid credit accounting and tenant display.
Three design decisions proved consequential:
- Measurement-side simplicity. True RMS computation on-device over 10-cycle windows, with per-point two-load calibration at install time (kettle + known resistive load). No voltage measurement per point — compound voltage is measured once at the hub and broadcast, trading per-point accuracy for cost.
- Outage-first persistence. Credit balances and cumulative energy live in a double-buffered flash journal committed on every 0.01 kWh increment and on brownout interrupt. No observed balance loss across 214 logged outages.
- Social legibility. The tenant display shows the same units (kWh and Kwacha) in the same format as ESCOM's own prepaid interface. Early prototypes showing percentages were consistently distrusted.
Results
Against a class-1 reference meter rotated across points monthly:
- Median absolute energy error: 2.2%; worst point-month: 3.1%.
- Error was dominated by the shared-voltage assumption during localized voltage sag (long feeder runs within the compound).
- Calibration drift over six months was below 0.5% for 10 of 12 points; the two outliers correlated with CT cores disturbed during unrelated wiring work.
- Billing disputes reported by the landlord fell from a baseline of 3.2/month to 0.4/month by study end.
Limitations and the certification gap
This system measures for information, not for legally binding billing — Malawi, like most jurisdictions, requires type-approved meters for revenue metering. In practice, all parties in the study treated the readings as authoritative because the alternative was not a certified meter but a heuristic. This regulatory gap — where the de facto competitor to an uncertified meter is no meter — deserves more attention in appropriate-technology literature than it receives.
Tamper behaviour was mild and creative: one household relocated a freezer to a neighbour's circuit on a rotating schedule. CT placement audits, not electronics, are the countermeasure.
Artifacts
Preprint PDFplaceholder — 18 pagesHardware design files and firmware are described in the associated project write-up on this site.