16 Energy-Aware Case Studies
Field Patterns, Failure Modes, Measured Ledgers, and Transferable Lessons
16.1 Start With the Evidence Record
A case study is useful when it tells you what was measured, what was assumed, and what changed after deployment. Without that evidence record, a battery-life claim is only a story.
Read each case by following the load ledger, source limits, environmental conditions, and validation gaps. The reusable lesson is the pattern, not the headline number.
Phoebe’s Why
A milliamp-hour is a count of charge, and charge alone does no work – it only becomes energy once it moves through a voltage. Two chemistries can quote different mAh numbers for reasons that have nothing to do with how much useful work they can deliver, because their voltage is set by which chemical reaction is running, not by how big the cell is. That is why this chapter’s own chemistry table can list more mAh for alkaline than for Li-ion and still be describing the weaker cell: comparing raw mAh across chemistries is like comparing prices quoted in different currencies without converting them.
The Derivation
Charge alone is not energy; energy requires the voltage it moved through:
\[E_{Wh} = V \times Q_{Ah}\]
For a chemistry with a roughly flat discharge plateau, the nominal voltage is a good stand-in for the whole discharge; for a chemistry with a sloped discharge curve, using the fresh open-circuit voltage overstates the energy actually delivered before cutoff.
Worked Numbers: This Chapter’s Own Chemistry Table
- Alkaline: \(1.5\text{ V}\times2.5\text{ Ah} = 3.75\) Wh
- Li-ion: \(3.7\text{ V}\times2.0\text{ Ah} = 7.4\) Wh – nearly double the alkaline energy despite 20% less mAh
- Li-SOCl2: \(3.6\text{ V}\times2.6\text{ Ah} = 9.36\) Wh
- CR2032: \(3.0\text{ V}\times0.22\text{ Ah} = 0.66\) Wh
- The trap: the alkaline cell’s 2500 mAh is \(2500/2000=1.25\times\) the Li-ion cell’s mAh rating, yet its actual energy is only \(3.75/7.4\approx0.507\times\) the Li-ion cell’s – about half. Alkaline’s own sloped discharge curve (voltage falls through the whole discharge rather than holding a plateau) makes the true gap larger still, because the 1.5 V nominal figure is the fresh-cell voltage, not the average delivered voltage. A design that racked and stacked cells by mAh alone would have picked the wrong chemistry for this chapter’s own 10-year, 15 uA sensor before self-discharge ever entered the comparison.
16.2 Energy-Aware Case Studies
This chapter uses anonymized field patterns rather than vendor stories. The goal is not to memorize product claims. The goal is to learn how to read an energy case study, identify the real cause of the improvement, and decide whether the lesson transfers to a new IoT design.
16.3 Learning Objectives
By the end of this chapter, you will be able to:
- Read an energy case study as an evidence record rather than a success story.
- Identify the dominant current state that drove each design change.
- Separate measured improvement from battery-life estimates and deployment assumptions.
- Compare communication, sensing, sleep, and board-leakage interventions.
- Explain why a case-study lesson may not transfer to another deployment.
- Build a compact case-study review for a battery-powered IoT design.
- A case study needs before and after ledgers with the same workload boundary.
- The largest improvement usually comes from the dominant energy consumer.
- A protocol swap is not automatically an energy win; payload, interval, coverage, and retry behavior matter.
- Deep sleep claims must include whole-board sleep current, not only MCU current.
- Long battery-life estimates need derating and field evidence.
- Transferable lessons are patterns, not copied component lists.
16.4 How to Read an Energy Case Study
Every case study in this chapter follows the same review map.
Ask five questions before accepting the lesson:
- What was measured? Current, voltage, time, radio retries, sensor warm-up, sleep leakage, and workload boundary.
- What dominated? Radio, sensing, compute, sleep leakage, regulator loss, or unexpected board current.
- What changed? Duty cycle, protocol, batching, sleep mode, sensor choice, power gating, or firmware state handling.
- What was re-measured? Same workload, same battery assumption, same service requirement.
- What transfers? The pattern that applies to another design, not the exact numbers.
16.5 Case Study 1: Field Soil Sensor
Initial symptom: A battery-powered soil sensor failed its seasonal-life target. The current trace showed that the radio connection sequence and idle loop dominated the daily budget.
Evidence
Baseline
After revision
Measurement interval
Sample and report every 15 minutes.
Sample every 15 minutes, report only when a threshold or scheduled batch required it.
Sleep state
MCU idle loop between readings; board current stayed high.
Deep sleep between readings, with sensor rail disabled.
Radio behavior
Full network reconnect for every report.
Shorter connection window and fewer reports.
Review lesson
The dominant cost was not the soil measurement.
Fix the idle and communication ledger before tuning the sensor.
Transferable pattern: When a field sensor fails a battery target, first inspect the full-period trace. If the trace never reaches a low-current sleep floor, communication and firmware-state fixes will usually beat sensor micro-optimizations.
16.6 Case Study 2: Asset Tracker
Initial symptom: An asset tracker met its average-current target in the lab but failed in field trials. The difference was poor coverage: retries and network attach time were longer than the design assumption.
16.6.1 Baseline assumption
Position report every fixed interval, with a short network attach and one successful upload.
16.6.2 Field evidence
Weak coverage created repeated attach attempts, longer transmit windows, and occasional queued reports.
16.6.3 Revision
Cache positions locally, reduce report frequency while stationary, and send batches when coverage is acceptable.
16.6.4 Service gate
The asset owner accepted delayed reports for stationary assets but required immediate escalation after movement.
Transferable pattern: Any tracker case study must include movement state, coverage state, retry behavior, and deadline class. A policy that works for stationary assets may be unacceptable for safety-critical moving assets.
16.7 Case Study 3: Industrial Vibration Monitor
Initial symptom: A vibration monitor consumed too much energy when it streamed raw samples to a gateway. The data movement cost exceeded the local computation cost.
Design choice
Why it helped
What had to be checked
Local feature extraction
Reduced upload size by sending compact spectral features instead of raw bursts.
The feature set still detected the target fault classes.
Adaptive interval
Measured slowly during stable operation, then increased frequency when vibration changed.
Escalation was fast enough for the maintenance use case.
Gateway batching
Reduced wireless overhead by grouping non-urgent summaries.
Buffered records survived resets and poor connectivity.
Transferable pattern: Do not assume local processing is worse for energy. When raw data is large and features are compact, local compute can reduce radio energy and improve battery life.
16.8 Case Study 4: Board Leakage Failure
Initial symptom: Firmware entered deep sleep correctly, but whole-board sleep current was far above the estimate. The MCU was not the problem; support circuitry stayed powered.
Regulator current The regulator quiescent current was larger than the MCU sleep current.
Debug bridge The USB or serial bridge remained powered in deployment builds.
Pullups and dividers Low-value resistors consumed continuous current across the battery.
Sensor standby The sensor looked asleep in firmware but its power rail remained enabled.
Indicator load Status LEDs and level shifters leaked through supposedly inactive paths.
Measurement fix The review switched from MCU-current estimates to whole-board current traces.
Transferable pattern: A low-power MCU does not make a low-power product. Sleep-state reviews must measure the whole board after the final hardware, firmware, and debug configuration are in place.
16.9 Pattern Map
The cases map to a small set of reusable energy lessons.
Pattern
Good evidence
Weak evidence
Duty-cycle repair
Current trace shows a real low-current sleep floor and bounded wake window.
Only firmware settings or datasheet sleep numbers are shown.
Communication repair
Upload size, attach time, retries, signal quality, and batching behavior are measured.
Only protocol names are compared.
Sensing repair
Accuracy, freshness, missed events, and sensor warm-up are checked.
Only lower sample rate is reported.
Board repair
Whole-board sleep current is measured after debug and peripheral shutdown.
Only MCU datasheet current is quoted.
16.10 Best-Practice Checklist
16.10.1 Start with the trace
Capture enough time to include wake, sense, compute, transmit, receive, retry, and sleep states.
16.10.2 Rank before optimizing
Optimize the largest measured contributor first unless a service or safety requirement says otherwise.
16.10.3 Preserve service
Battery gains are not valid if they miss required events, stale the data, or remove necessary feedback.
16.10.4 Include overhead
Batching, compression, inference, and scheduling consume energy and memory. Include them in the revised ledger.
16.10.5 Derate estimates
Battery chemistry, temperature, aging, self-discharge, voltage cutoff, and field retries reduce ideal estimates.
16.10.6 Record transfer limits
Document where the lesson applies and where it does not: coverage, payload size, deadline, and maintenance model.
16.11 Common Pitfalls
16.11.1 Copying numbers
Case-study currents and lifetimes are not reusable design constants. Reuse the measurement method.
16.11.2 Ignoring field coverage
Lab radio energy can be much lower than field energy when retries and attach failures appear.
16.11.3 Hiding service loss
Longer sleep or batching is not a win if it violates freshness, safety, or user-feedback requirements.
16.11.4 Datasheet-only sleep
Whole-board leakage can dominate even when the MCU reaches its advertised sleep current.
16.11.5 Treating estimates as measurements
Battery-life spreadsheets are useful, but promotion requires traces from the real firmware and board.
16.11.6 No rollback path
Adaptive field policies need fallback behavior when context, coverage, or sensing quality degrades.
16.13 Check Your Understanding
16.14 Knowledge Check: Case-Study Evidence
16.15 Knowledge Check: Transferable Lesson
16.16 Matching Quiz: Case Patterns
16.17 Ordering Quiz: Case-Study Review
16.18 Label the Diagram: Case-Study Evidence
16.19 What’s Next
Connect case-study evidence to battery, regulator, and deployment design assumptions.
Practice collecting and interpreting an energy trace.
Energy-Aware Low-Power Strategies
Choose practical sleep, wake, and low-power strategies after reading the case patterns.
Energy-Aware Interactive Tools
Use calculators and tools after the evidence pattern is clear.
16.20 The Cell Discharges Even When The Sensor Sleeps
Across the case studies in this chapter, one factor separates the multi-year successes from the disappointments: battery chemistry. A cell is not just a capacity number. Each chemistry has its own self-discharge rate, voltage, temperature range, and pulse-current behavior, and those properties decide whether a design that looks fine on the bench survives years in the field.
For a very low-power device, the surprising result is that the chemistry, not the workload, can set the lifetime. A cell that quietly loses 2% of its charge every month can empty itself faster than a microamp-scale sensor ever could, so choosing the wrong chemistry wastes most of the capacity before the load touches it.
Intuition only: self-discharge is a leak that runs in parallel with your load, all the time, whether or not the device is awake. Compare that leak against your average current before trusting a capacity figure.
For a 10-year, 15 uA soil sensor, the load itself needs about 15 uA x 87660 h = 1315 mAh before derating. A 2000 mAh Li-ion cell losing 2% per month leaks about 40 mAh each month, or 480 mAh per year; over 10 years the calendar loss alone is more than twice the cell's nominal capacity. A 2600 mAh Li-SOCl2 cell at about 1% per year loses about 26 mAh per year, or 260 mAh over the same period, leaving enough charge for the load plus reserve. The same average-current design therefore succeeds or fails before firmware changes, simply because the chemistry leak is different.
Chemistry At A Glance
Li-SOCl2
3.6 V, very low self-discharge (about 1%/year), wide temperature, but limited continuous current. The default for multi-year unattended sensors.
Alkaline
1.5 V, cheap, moderate self-discharge (a few %/year), but weak in cold and under pulse loads, and prone to leakage over long deployments.
Li-ion / LiPo
3.7 V, rechargeable, energy-dense, but self-discharges roughly 2-3% per month and cannot be charged below 0 C. Good only if recharged or serviced.
Coin cell (CR2032)
3 V, about 220 mAh, low self-discharge, but high internal resistance limits pulse current. Fine for tiny loads, risky for radio bursts.
Overview Knowledge Check
16.21 Add Self-Discharge As A Parallel Current
A useful first-order model treats average self-discharge as a constant current draining the cell alongside the load. Convert the rated percentage into an equivalent current: I_self = (rate x capacity) / period. Then lifetime becomes life = usable capacity / (I_load + I_self). When I_self approaches or exceeds I_load, the chemistry, not the workload, governs the outcome.
Worked Example: A 15 uA, 10-Year Sensor
Compare two chemistries for the same 15 uA average load and a 10-year target.
- Li-SOCl2, 2600 mAh, about 1%/year: self-discharge = 26 mAh/year, and 26 mAh / 8766 h = about 3 uA equivalent. Total draw = 15 + 3 = 18 uA. With ~2340 mAh usable, life = 2340 / 0.018 = 130000 h, about 14.8 years. Meets the target with margin.
- Li-ion, 2000 mAh, about 2%/month: self-discharge = 40 mAh/month, and 40 mAh / 730 h = about 55 uA equivalent. Total draw = 15 + 55 = 70 uA. Even at a generous 2000 mAh usable, life = 2000 / 0.070 = 28600 h, about 3.3 years. Fails the target - dominated by self-discharge.
The Li-ion cell holds less than the Li-SOCl2 cell here, but that is not why it loses. It loses because its self-discharge current is nearly four times the load, so most of its charge never reaches the sensor.
Run the same calculation whenever a case study changes chemistry. First convert the chemistry leak into current, then add it to the measured workload current, then apply a usable-capacity derating. If the project requires 20% end-of-life reserve, the Li-SOCl2 example should use about 2080 mAh instead of 2600 mAh; at 18 uA total draw that still gives about 13.2 years. The Li-ion example remains below four years even before cold, ageing, or safety reserve are considered.
Equivalent Self-Discharge Current
Practitioner Knowledge Check
16.22 Temperature Moves Every Number
Chemistry ratings are quoted at room temperature, but field enclosures rarely sit at room temperature. Self-discharge is a chemical reaction, and its rate rises roughly with temperature - as a rule of thumb it can nearly double for every 10 C increase. A sealed enclosure baking on a sunny rooftop can therefore run at twice the self-discharge current you budgeted from the datasheet, quietly halving the margin computed on the bench.
Temperature also shifts usable capacity and safety limits per chemistry. Alkaline cells lose a large fraction of their capacity in the cold, so a winter deployment sees far less than the rated mAh. Li-ion cells discharge in the cold but must not be charged below 0 C, because charging then plates lithium and permanently damages the cell - a hard constraint for solar-recharged nodes in cold climates. Li-SOCl2 cells tolerate very wide temperatures but suffer passivation after long storage or cold soaking, which delays the voltage recovery under a sudden pulse. None of these effects appear in a single room-temperature capacity figure.
Put numbers on the derating. If the room-temperature Li-SOCl2 leak is 3 uA at 20 C and the enclosure spends summer near 40 C, two 10 C steps can make the leak roughly 3 uA x 2 x 2 = 12 uA. The same 15 uA sensor now budgets 27 uA average before radio-pulse margin. With 2340 mAh usable, the lifetime falls to 2340 mAh / 0.027 mA = 86667 h, or about 9.9 years. A design that looked comfortably above 10 years at room temperature is suddenly marginal in the actual enclosure.
Temperature Effects To Budget
Faster self-discharge
Roughly doubles per 10 C rise. A hot enclosure can erase the self-discharge margin you planned.
Cold capacity loss
Alkaline can deliver a fraction of its rated capacity below freezing. Rate the pack at the deployment temperature.
Li-ion charge limit
No charging below 0 C without plating and damage. Cold solar nodes need a charge-temperature gate or a different chemistry.
Passivation
Li-SOCl2 can show a voltage delay after storage or cold, weakening the first pulse until the film clears.
Under-the-Hood Knowledge Check
16.23 Summary
This chapter uses case studies to show how energy-aware design changes across deployments. It compares sensing schedules, communication choices, harvesting assumptions, maintenance constraints, and operational tradeoffs.
16.24 Key Takeaway
Case studies show that energy design is situational. Battery size, duty cycle, radio coverage, maintenance access, data freshness, and environmental conditions decide which strategy is defensible.