12 Energy Harvesting: Source Reality and Field Proof

12.1 Start With the Situation

A bench ledger balances and the storage calculation looks safe, but the field source is weaker and less regular than its peak rating. The team must test the actual light, temperature difference, motion, or RF environment and record what the node survives.

12.2 Overview

This route applies reality checks to non-ideal sources, then builds the field record, failure tests, and evidence needed for an honest autonomy claim.

This is part 2 of 2. Review Energy Harvesting: Budgets, Conversion, and Storage when you need the first route.

12.3 Learning Objectives

By the end of this chapter, you will be able to:

• evaluate indoor solar, thermal, vibration, and RF source limits
• design a field verification record for harvested energy
• reject perpetual-operation claims that lack measured seasonal evidence

12.4 Chapter Roadmap

Follow the original sections below in order. They begin at the reviewed split boundary and keep every worked example, figure, check, and supporting banner with the section that owns it.

12.5 Indoor and Non-Solar Reality Checks

Harvesting proposals often fail because the source was described qualitatively instead of measured.

TEGs Need Delta-T Across the Module

A hot pipe or warm enclosure is not enough. Thermoelectric output depends on the temperature difference across the module after thermal contact resistance and heatsink limits are included.

The TEG decision turns on temperature difference across two faces, not on ambient temperature alone. The photograph in Figure 12.1 makes that physical boundary concrete before the chapter evaluates whether a site can sustain useful thermoelectric power.

On Figure 12.1, read the etched Hot Side marking first and then distinguish the red and black electrical leads leaving the module. “Hot Side” identifies one thermal face; it does not guarantee that the opposite face stays colder, and the leads merely carry whatever electrical output the maintained gradient produces. The deployment question therefore returns to heat-source coupling, heat rejection, and measured delta-T rather than the module’s presence.

12.6 Field Verification Record

Before claiming energy-neutral operation, record evidence that can be retested.

12.7 Common Pitfalls

12.8 Knowledge Check

12.9 Quiz: Energy-Neutral Budget

12.10 Matching Quiz: Source to Design Gate

12.11 Ordering Quiz: Harvesting Design Review

12.12 Label the Diagram: Harvesting Energy Path

12.13 Related Chapters

12.18 What’s Next

12.22 Harvest Is A Trickle, So Budget Averages And Store The Rest

Energy harvesting replaces or supplements a battery by pulling power from the environment. The catch is scale: ambient sources deliver a trickle compared with a battery's reserve, and that trickle is intermittent. Designing a harvesting node is therefore an averages-and-storage problem, not a peak-power one. You compare the average power you can harvest against the average power the device consumes, and you add storage to ride through the gaps when the source is absent.

The sources differ by orders of magnitude, and knowing the rough numbers keeps a design honest. Representative electrical output densities are about 15 mW per square centimeter for outdoor solar, only about 10 microwatts per square centimeter for indoor solar, tens of microwatts per square centimeter for a body-heat thermoelectric generator, around 100 microwatts per cubic centimeter for machine-vibration piezo, and well under 1 microwatt per square centimeter for ambient radio-frequency energy.

A quick budget exposes the difference. A 0.5 mW sensor needs 0.5 x 24 = 12 mWh/day. One square centimeter of outdoor solar with 2 peak-sun-hours and a 0.7 derating can produce about 15 mW x 2 h x 0.7 = 21 mWh/day, so the daily energy balance closes. Indoors, the same square centimeter at 10 uW for 10 lit hours and 0.7 derating produces only 0.07 mWh/day; the same load would need about 12 / 0.07 = 171 cm2 before storage losses. The source choice changes the physical design, not just the BOM line.

12.23 Source Reality

12.24 Overview Knowledge Check

12.25 Size For Energy-Neutral, Not Peak

Convert the intermittent source into a daily energy using peak-sun-hours (PSH), the equivalent hours per day at full rating. Then require panel_daily_energy x derating >= load_daily_energy and size storage for the longest expected gap.

12.26 Worked Example: Outdoor Solar Sensor

The device averages 0.2 mA at 3.3 V, which is 0.66 mW, or 15.8 mWh per day. Use a conservative 2 peak-sun-hours per day for reliability in poor weather and a system derating of 0.7 for conversion, dirt, and angle losses.

• Panel size: required panel peak power = 15.8 mWh / (2 h x 0.7) = 11.3 mW. At 15 mW/cm2 outdoors, that is only about 0.75 cm2 of cell - a postage stamp comfortably powers this load.
• Storage for autonomy: to survive 3 days with no harvest, store 15.8 mWh x 3 = 47.5 mWh. At 3.7 V that is about 12.8 mAh, met by a small 20-50 mAh lithium cell.
• Result: the outdoor design is generous; the panel is tiny and a small cell provides multi-day ride-through.

Notice the two independent sizes: the panel is sized by the daily energy balance, and the storage is sized by the worst-case gap. Getting one right does not excuse the other.

Then add engineering margin before calling it viable. If the site spends one winter week at only 1 peak-sun-hour, the same panel harvests 11.3 mW x 1 h x 0.7 = 7.9 mWh/day, half the load. A controller can respond by stretching the reporting interval, disabling a high-current sensor, or declaring a maintenance fault before the buffer reaches brownout. The harvesting budget therefore needs both a nominal sizing line and a degraded-mode policy line.

12.27 Harvesting Sizing Ledger

12.28 Practitioner Knowledge Check

12.29 Indoor Is A Thousand Times Weaker, And Peak Is Not Average

The most common harvesting mistake is moving an outdoor-proven solar node indoors. Outdoor cell output is around 15 mW/cm2; typical office lighting yields only about 10 uW/cm2 - roughly a thousandfold drop. Take the same 15.8 mWh/day load. Indoors, with light available perhaps 10 hours a day, a square centimeter gathers about 10 uW x 10 h = 0.1 mWh per day. Meeting the load after derating needs about 22.6 mWh/day, which demands roughly 226 cm2 of indoor cell - a sheet the size of a page - versus under 1 cm2 outdoors. The panel that trivially powers the outdoor node cannot power the indoor one.

Two observations govern this part of indoor is a thousand times weaker, and peak is not average: Laser Light and Sunny Day. Their arrangement in the diagram at Figure 12.2 reveals whether the proposed boundary is complete.

Compare Laser Light and Sunny Day first in the illustration at Figure 12.2: one adds a distinct review condition, while the other adds a distinct review condition. Only then bring in Soccer Stadium, which adds a distinct review condition. This three-part reading supports the stated result: Solar power scale comparing laser light, sunny day, stadium light, overcast day, corridor light, street light, and candle light for energy harvesting. It is the chapter’s bridge back to indoor is a thousand times weaker, and peak is not average.

The second trap is confusing peak power with average power. A cell rated 15 mW/cm2 delivers that only in full sun. Rating a node on the peak overstates the daily harvest by the ratio of a full day to the actual peak-sun-hours, and the shortfall is worst in the season you can least afford it. Storage type follows from the gap length: for a few hours of ride-through a supercapacitor works and tolerates endless cycles, but for days of autonomy the energy needed forces a battery, because a supercapacitor large enough would be impractical.

Buffer math uses usable energy, not nameplate energy. A 50 mAh lithium buffer at 3.7 V stores about 50 x 3.7 = 185 mWh, but an 80% depth-of-discharge limit leaves 148 mWh. That covers the 15.8 mWh/day example for about 148 / 15.8 = 9.4 days before converter quiescent current, cold derating, and aging. A 10 uA always-on charger path at 3.7 V consumes another 0.010 mA x 3.7 V x 24 h = 0.89 mWh/day, which is small outdoors but material indoors.

12.30 Reality Checks Before Committing

12.31 Under-the-Hood Knowledge Check

12.32 Energy-Arrival And Random-Access Calculator

An energy-harvested radio cannot schedule from average power alone. Energy arrives unevenly, storage is finite, sleep electronics consume part of every arrival, and a random-access collision can force another expensive transmission. The operational question is therefore: given the current buffer and a conservative arrival rate, when may this node attempt a transmission without stealing energy from its survival reserve?

The calculator uses expected values to expose the policy. Its slotted random-access model assumes each of the contending nodes transmits independently with probability $p$ in a slot, giving this node a success probability of $p(1-p)^{n-1}$. Real protocols add carrier sensing, capture, acknowledgements, retry limits, and time-varying traffic; replace the teaching model with a measured trace before deployment.

Try three cases: one node with predictable harvest, many nodes with the same access probability, and a low-harvest interval with a nearly empty buffer. Observe why reducing traffic can improve both collision probability and energy survival. The optional graduate path is to replace the expected-value policy with a stochastic arrival trace and compare online policies; the practitioner path is to measure the source, charger quiescent current, buffer limits, packet energy, success rate, and missed-deadline behavior at the real site.

12.33 Energy-Harvesting AWGN Capacity

The mean arrival is an available power budget only when storage and policy keep that energy from spilling. Figure 12.3 pairs the infinite- and finite-battery cases with their corresponding AWGN expressions.

In Figure 12.3, Infinite battery can shift surplus energy to later slots, supporting P∞ = E[Y] under causality. Finite battery Bmax clips arrivals above free capacity, so Finite-store bound uses Puse = E[Y] − E[E_spill]; that is precisely when the mean alone overstates the usable transmit budget.

The finite-store ledger makes that clipping mechanism inspectable slot by slot. Begin with the 10 mJ store, then change only Harvest scale from 1.00 to 1.50: the arrival total grows, but the extra burst spills before later service can use it.

An energy-harvesting transmitter differs from an ordinary battery-powered radio because its transmit power is constrained by causality: energy cannot be spent before it arrives. Let $E_t$ be the energy harvested in slot $t$, $P_t$ the transmit energy used in that slot, and $B_t$ the energy available at its start. With storage capacity $B_{max}$, the state evolves as

The inequality is the causality rule; the min is the overflow rule. Energy arriving when the store is full is clipped and cannot be recovered later.

For an additive white Gaussian noise channel of bandwidth $W$ hertz, one-sided noise spectral density $N_0$, and constant average signal power $P$, Shannon capacity is

With ideal, lossless, unbounded storage and a stationary ergodic arrival process of mean harvested power $\bar P_h=\mathbb{E}[E_t]/T_s$, sufficiently long coding and an energy-causal save-then-transmit policy can approach the ordinary AWGN result with $P=\bar P_h$:

Why does only the mean appear? An infinite store can absorb high-arrival slots and release their energy during low-arrival slots. Over a long horizon, the buffer smooths the random source into an average-power constraint. This conclusion depends on the assumptions above; finite horizon, leakage, conversion loss, non-stationary weather, or strict delay limits can all make the arrival shape matter again.

Finite storage changes the problem. A single arrival is admitted as

so the usable mean can fall below the harvested mean. A rough empty-buffer clipping approximation replaces arrivals by $\min(E_t,B_{max})$, but

is not a general finite-battery capacity formula. Actual overflow depends on the current buffer and power-control policy, and the logarithm is concave, so bursty power allocation does not behave like constant average power. Capacity is obtained by optimizing an energy-causal policy over the buffer state and arrival statistics, commonly through dynamic programming or Markov decision methods after discretization.

This produces three engineering regimes:

• If $B_{max}$ comfortably spans the largest relevant energy drought and burst, the infinite-buffer approximation can be useful.
• If arrivals frequently overflow the store, increasing capacity or transmitting opportunistically can recover otherwise discarded energy.
• If the store is often empty, the policy must reserve energy for deadlines or channel opportunities; the mean arrival alone is misleading.

The design sequence is therefore: characterize the arrival distribution and correlation; measure conversion and leakage losses; choose a candidate $B_{max}$; simulate the state recursion with real traces; optimize or compare causal policies; then report throughput, outage, overflow, and missed-deadline rates together.

12.34 Solar Density and the Converter Contract

A useful first-pass light-harvesting anchor is about 1 $\mu$W/mm$^2$ under a reasonably bright condition. Because $1\ \text{cm}^2=100\ \text{mm}^2$, a 3 cm$^2$ cell at that density produces roughly

After 70% end-to-end conversion, about 210 $\mu$W reaches storage or load. A node averaging 100 $\mu$W could be energy-neutral while that lighting persists and could bank the surplus for dark periods. “Indefinite life” therefore means the long-term harvested energy covers load, losses, storage leakage, and degradation; it does not mean a few square centimetres work in every room or season.

The power-management circuit must satisfy three requirements at once:

Quiescent current belongs in the same ledger. A controller drawing 5 $\mu$A from a 1 V harvesting input consumes 5 $\mu$W continuously—5% of a 100 $\mu$W source before useful work begins.

12.35 Series-Parallel Harvesting Converter

A converter block becomes reviewable when its internal feedback and release gates are visible. Figure 12.4 opens the path from weak ambient sources to a regulated, power-good load rail.

In Figure 12.4, Rectify and accumulate includes the cold-start condition before a regulated controller supply exists. Switch-capacitor stages use non-overlap φ1 / φ2 clocks, Sense and choose gain prevents source collapse, and Qualify the load rail uses hysteretic power-good rather than exposing the load to a chattering undervoltage rail.

A switched-capacitor converter changes voltage by reconfiguring flying capacitors between charge and transfer phases. In one phase capacitors charge in a series or parallel arrangement; in the other, switches reconnect them so their voltages add, divide, or transfer charge to the output. A family of switch configurations provides discrete conversion ratios. An eight-ratio design can therefore cover a 0.6–2.4 V source while regulating a roughly 1.2 V load more efficiently than one fixed ratio across the whole range.

Follow the control loop step by step:

1. A source such as light, vibration, a temperature difference, or rectified RF charges an input reservoir. Its series resistance is part of the source model.
2. A reference and feedback divider compare $V_{out}$ with the target rail.
3. A gain controller selects one of the available capacitor ratios so the ideal converted voltage sits near the target rather than forcing regulation through avoidable loss.
4. A non-overlap clock generator creates phases $\phi_1$ and $\phi_2$ that are never high together. This prevents switches from momentarily shorting a charged capacitor or supply node.
5. The switch matrix moves charge through the selected series-parallel topology.
6. A current sensor and oscillator adjust switching frequency: more frequent packets of charge support load, while a weak source or light load needs a lower rate to avoid switching loss.
7. Power-good logic releases the load only after the storage/output node is usable and can return it to reset before an uncontrolled brownout.

The converter does not create energy. For ideal ratio $M$, $V_{out}\approx M V_{in}$, while real output is bounded by efficiency:

Ratio changes, bottom-plate parasitics, switch resistance, clock generation, leakage, and charge redistribution all reduce $\eta$. This is why ratio selection and frequency control must be measured over source voltage and load current, not validated at one nominal point.

12.36 Step-Charging Derivation

Connecting an ideal voltage source directly to an initially discharged capacitor stores $\tfrac12CV^2$ in the capacitor and dissipates the same amount in the series path. The one-jump redistribution loss is

If the voltage rises through intermediate levels $V_0,V_1,\ldots,V_N=V_f$, the redistribution loss becomes

For $N$ equal steps, $\Delta V=V_f/N$, so

The inverse-$N$ result is the key: five equal steps ideally reduce charge-redistribution loss to one fifth of the single jump.

Take $C=100$ nF and $V_f=1.3$ V. The correct one-step value from the printed parameters is

Five equal steps give $84.5/5=16.9$ nJ, close to an 18 nJ measured or rounded result. A source graphic associates these same $C$ and $V_f$ labels with 192 nJ, but those numbers do not satisfy $\tfrac12CV^2$; the chapter keeps the physically consistent derivation instead of propagating that mismatch. Real step charging also spends energy in switch gates, clock generation, leakage, and finite resistance, so ever-smaller steps eventually lose to control overhead.

12.37 Radio Energy per Bit: Read the Denominator

Energy per bit is power divided by useful bit rate:

One historical comparison illustrates how strongly the denominator and link assumptions matter:

The arithmetic checks—for example, $0.147$ mW$/960$ bit/s $=153$ nJ/bit—but the table is not a technology ranking. Payload rate, range, receiver state, startup, association, acknowledgement, retry, idle listening, and tail time differ. A high-rate radio can amortize large instantaneous power over many bits once connected, while sending one tiny report may be dominated by its setup energy.

For a real harvester, calculate session energy first:

then divide by successfully delivered application bits, not nominal PHY bits. This links the converter and radio designs: storage and power-good thresholds must support peak session current even when the long-term harvested average is sufficient.

12.38 Summary

This chapter introduces energy harvesting sources such as solar, vibration, thermal, and RF. It connects harvested power, storage, load profiles, energy-neutral operation, and reliability during low-input periods.

12.39 Key Takeaway

Energy harvesting works only when the long-term energy balance closes. Size storage for gaps, measure realistic source availability, and design graceful behavior for periods when harvested power is below demand.