A battery-powered node has a clear expiration date: the energy runs out, the operation stops. A node with Energy Harvesting has no battery to replace, but has a subtler problem - the energy never arrives at a constant rate, and the system must be designed to survive exactly those intervals when the source offers nothing. This lecture covers the whole chain: the available sources (solar, thermoelectric, vibration, RF), power conversion and management, state-based consumption modeling, and sizing the storage for the worst case, not the average.
1The subject and structure of the lecture6 min
Many embedded systems must operate for a long time in locations where mains power is not available, and replacing batteries is difficult or costly - distributed IoT nodes, industrial sensors on machinery, structural monitoring systems, wearable medical devices. Lower consumption of microcontrollers and sensors has made it possible to power such systems directly from the energy available in the environment - a process called Energy Harvesting.
- Average power and the energy of the autonomy are calculated from the state-based consumption profile
- Adaptive energy management adjusts the frequency of operations based on the available energy
In many applications, the harvested energy supplements stored energy and reduces the frequency of maintenance visits. Only under well-defined conditions does one achieve fully autonomous operation over the very long term.
Learning outcomes
- Describe the complete energy chain, from source to embedded system
- Calculate a battery's autonomy and the energy-neutrality condition
- Estimate the power available from photovoltaic, thermoelectric, vibration and RF sources
- Size a source and a storage reserve for the worst-case scenario
- Model the state-based consumption of a node and calculate the average power and energy
- Explain the role of energy thresholds in intermittent operation
2The general principle of Energy Harvesting8 min
An Energy Harvesting system converts a form of energy available in the environment into a voltage and current usable by the electronic circuit. The general energy chain is:
source → converter → power management → storage → embedded
systemThe source supplies power that depends on the outside environment. The converter performs the physical transformation (light into electrical energy, temperature difference into voltage, mechanical deformation into electrical charge). Power management adapts the voltage, tracks the source's optimal operating point and controls the storage charging.
P_useful = η_conv · η_PMIC · P_source - the overall efficiency
is not constant, it varies with input voltage, load current and temperature.Why storage is needed
The harvested energy and the consumed energy are usually not available at the same moment - a photovoltaic panel produces during the day, but the node must also work at night; a transceiver may require, for a few milliseconds, much more power than the source is instantaneously providing.
E_s(t₂) = E_s(t₁) + ∫[P_h(t) - P_c(t) - P_l(t)]dt, with
0 ≤ E_s(t) ≤ E_max3Autonomy and energy neutrality10 min
Autonomy represents the period during which the system can operate without external power or maintenance intervention. For a system powered exclusively by a battery, the autonomy is limited by the initial energy available.
T_ideal = C_B / I_average, where C_B is the capacity in
amp-hours.A 2000 mAh battery powers a system with an average current of 0.1 mA. What is the ideal autonomy, and the real one, with a utilization factor k_u = 0.75?
See the solution
T_ideal = 2000/0.1 = 20 000 h ≈ 2.28 years
This is a theoretical value. The real autonomy, affected by self-discharge, temperature,
aging and the minimum accepted voltage, is better approximated by
T_real ≈ k_u × C_B/I_average: with k_u = 0.75, T_real ≈ 1.71 years. In a
network with thousands of nodes, periodically replacing batteries can become more costly than the
monitored equipment itself - exactly the motivation for Energy Harvesting.
P̄_h ≥ P̄_c + P̄_l (the average harvested power covers the consumption
plus the losses).The average condition is not enough if harvesting and consumption are distributed differently over time - a system can produce enough energy over a month and still shut down during a consecutive multi-day period without a source. A robust design must analyze an explicit worst-case scenario: the month with the lowest production, several consecutive days with reduced light, temperatures that reduce battery capacity - not just the annual average.
When the available energy drops, the system can lower the level of service provided before reaching shutdown - adaptive energy management, with modes decided by the storage level:
mode = normal (E_s ≥ E_H) | economy (E_L < E_s < E_H) |
survival (E_s ≤ E_L)In economy mode, the system may transmit less often; in survival mode, it may keep only the essential measurements and stop non-critical communications.
4Energy sources for embedded systems8 min
Choosing the source starts by measuring the actual conditions in the operating environment - a technology that performs well in the lab can deliver very little energy if it is not matched to the application.
| Source | Converter | Main advantage | Main limitation |
|---|---|---|---|
| Light | Photovoltaic cell | Relatively high power | Depends on illumination |
| Thermal gradient | Thermoelectric generator | No moving parts | Needs a temperature difference |
| Vibration | Piezoelectric / electromagnetic / capacitive | Suited to machinery | Sensitive to frequency and amplitude |
| Radio frequency | Antenna and rectifier | Contactless operation | Low ambient power |
| Human motion | Mechanical converter | Integration into wearables | Availability depends on the user |
A source must be evaluated both by power and by availability: solar energy has estimable daily and seasonal variation; industrial vibration may only be available while the machinery is running; thermoelectric energy vanishes if the two faces of the converter reach the same temperature. A source with high average power but long interruptions may need more storage than a modest but continuously available source.
E_h,min ≥ E_c,max + E_reserve, for the worst-case scenario. If a
single source does not satisfy the condition, sources can be combined: P_h(t) = Σ
P_h,j(t) (hybrid architecture, for example photovoltaic + thermoelectric).Using several sources increases availability, but introduces more converters, extra matching circuits, more complex control and higher cost and size - the decision must be made by comparing the extra energy gained with the resources needed for the integration.
5Photovoltaic conversion9 min
Photovoltaic conversion is one of the most widely used methods of powering autonomous embedded systems - the cells convert light energy directly into electrical energy, with no moving mechanical parts.
P_PV = G·A·η_PV, where G is the incident irradiance, A the
active area, η_PV the conversion efficiency - useful for a preliminary sizing, but the
efficiency varies with temperature, the light spectrum and the operating point.A photovoltaic cell does not behave like an ideal voltage source: at open terminals
(I=0, V=V_OC) and short circuit (V=0, I=I_SC), the delivered power is
zero in both cases. Maximum power is obtained at an intermediate point, called the MPP
(Maximum Power Point): P_max = V_MPP · I_MPP.
FF = (V_MPP · I_MPP) / (V_OC · I_SC) - gives an indication of the shape of the
characteristic and the electrical performance of the cell. Connecting the panel directly to a
battery or a load does not guarantee operation at the maximum-power point - which is why an
MPPT (Maximum Power Point Tracker) circuit is needed, which changes the apparent impedance
seen by the panel, tracking the condition dP_PV/dV_PV = 0, through methods such as
perturb-and-observe or maintaining a fraction of V_OC.P_PV,min = E_consumption,day / (H_equiv,min · η_system), where
H_equiv,min is the number of equivalent hours at nominal power, in the worst-case scenario.A node consumes 24 mWh/day. What minimum nominal power must the panel have, if in the worst-case scenario H_equiv,min = 2 h and η_system = 0.70?
See the solution
P_PV,min = 24 mWh / (2 h × 0.70) ≈ 17.1 mW
In practice a margin is added for shading, soiling, aging and weather variation - a panel two or three times larger than the minimum value can be justified, but should not be chosen without an explicit energy analysis showing why.
6Thermoelectric and mechanical conversion10 min
Thermal and mechanical energy are important sources in industrial environments - pipes, motors, compressors and moving structures can supply local energy for sensors.
The thermoelectric generator
Thermoelectric conversion uses the Seebeck effect: a temperature difference produces an electrical voltage.
V_OC = S_equiv·ΔT, with ΔT = T_H - T_C. Power delivered to a
load R_L: P_L = V_OC²R_L/(R_i+R_L)², maximum at R_L = R_i, where
P_max = V_OC²/(4R_i).If the cold side cannot dissipate the heat, the temperature difference at the module's terminals drops, even if the ambient temperatures look favorable - a calculation that ignores the real heat dissipation systematically overestimates the available power.
Vibration conversion
A vibration converter is often represented by a mass-spring-damper system:
mẍ + cẋ + kx = -mÿ, with undamped natural frequency
f₀ = (1/2π)√(k/m). Near resonance, the converted energy can increase
significantly - but a very narrow resonance becomes disadvantageous if the source frequency
changes.
A motor runs at n = 1500 rpm. The fundamental rotation frequency is
f_rot = n/60 = 25 Hz - but the spectrum can contain harmonics and frequencies
associated with the bearings, not just this fundamental component.
Three main conversion mechanisms: piezoelectric (deforming the material produces
electrical charge), electromagnetic (relative magnet-coil motion induces a voltage,
v(t) = -N·dΦ/dt) and electrostatic (the vibration changes a capacitance,
C(x) = εA/d(x), usually requiring an initial bias voltage or an electret material).
Increasing the useful bandwidth usually reduces the maximum amplitude obtained at a single
frequency - the design is a trade-off between peak power and robustness to source variation.
7RF energy harvesting8 min
An RF Energy Harvesting system captures electromagnetic energy through an antenna and converts it into a DC voltage. A distinction must be made between ambient harvesting (the power depends on already-existing transmitters, varying without control) and dedicated transfer (the transmitter, the antennas and the distance are specifically designed to power the receiver).
P_R = P_T·G_T·G_R·(λ/4πd)², or logarithmically:
P_R[dBm] = P_T[dBm] + G_T[dBi] + G_R[dBi] - L_path[dB]. It does not fully describe
propagation indoors or in the near field.The antenna + rectifier assembly is called a rectenna: antenna → impedance matching →
RF-DC rectifier → filter → storage/DC-DC. The resulting DC power is
P_DC = η_RF-DC · P_R, and the rectifier's efficiency drops significantly at very low
power levels - efficiency measured at high power cannot be automatically extrapolated to much
weaker ambient RF energy.
A measurement and a transmission need E_operation = 1 mJ. The average DC power obtained from ambient RF is P_DC,average = 10 µW. What is the ideal minimum charging time between two operations?
See the solution
T_charge ≥ E_operation / P_DC,average = 1 mJ / 10 µW = 100 s
In practice the losses, the idle consumption and the minimum energy needed to start up must be included - the real value will be larger than 100 s.
Ambient RF energy is suited to slow accumulation, intermittent operation and very low-consumption sensors. Dedicated transfer is used in passive RFID, NFC and reader-interrogated sensors - a passive RFID device does not transmit by actively generating its own wave, but instead changes the antenna's impedance and modulates the reflected signal, drastically reducing the energy needed for communication.
8Power conversion and management9 min
Energy-harvesting sources generally cannot directly power an embedded system - the voltage and current vary with illumination, temperature or vibration, while the microcontroller needs a stable voltage. The intermediate circuit is called the PMIC (Power Management Integrated Circuit): rectification, matching the operating point, DC-DC conversion, charge control, storage protection.
η_overall = η_rectification × η_DC-DC × η_charging × η_regulator - even if each stage
has a high individual efficiency (for example 90% each), the product of four such stages drops
below 66% - which is why every link in the chain must be optimized, not just one.Cold start
At the beginning, the storage element is discharged, and the PMIC must start up using only the power the source is delivering instantaneously - a difficulty specific to Energy Harvesting systems, with no equivalent in a system powered by a pre-charged battery. The usual solution: a start-up path (very low own consumption, works from very low voltages) and a main path, with higher efficiency, activated once the storage has accumulated enough energy.
MPPT and impedance matching
The MPPT circuit changes the apparent impedance presented to the source, Z_in =
V_s/I_s, tracking the maximum power transfer. For a Thévenin-modeled generator, the
condition is R_load = R_internal. The MPPT must not be evaluated only by the
conversion efficiency - the own consumption of the measurement and control circuit must also be
included: P_gain = P_MPPT - P_without_MPPT - P_control. For very weak sources, a
simplified algorithm may produce more net energy than a complex control scheme, if the
latter consumes more than it gains.
Energy thresholds and intermittent operation
Some systems do not harvest enough power for continuous operation - energy accumulates up to a
threshold, the system performs an operation, then returns to charging:
charge → start-up → measurement → transmission → shutdown, with two distinct
thresholds (V_start > V_stop) to avoid repeated switching at the boundary.
E_usable = ½C(V_start² - V_stop²). The operation can be executed
only if E_usable ≥ E_operation + E_reserve.Important data must be saved before shutdown, and operations on non-volatile memory must be designed so that an unexpected shutdown in the middle of a write does not leave the system in an inconsistent state - a software design concern directly tied to the unpredictable nature of the energy source.
9State-based consumption modeling10 min
A sensor node typically operates in several states: sleep, wake-up, measurement, processing,
transmission, back to sleep. The energy consumed in a state is Eᵢ = Pᵢtᵢ, and the
energy of a complete cycle is E_cycle = Σ Pᵢtᵢ.
A node has a cycle period T_C = 600 s, with the profile:
| State | Power | Duration | Energy |
|---|---|---|---|
| Sleep | 8 µW | 598 s | 4.784 mJ |
| Measurement | 5 mW | 1 s | 5 mJ |
| Processing | 12 mW | 0.5 s | 6 mJ |
| Transmission | 120 mW | 0.5 s | 60 mJ |
What is the average power, and the daily energy?
See the solution
E_cycle = 4.784 + 5 + 6 + 60 = 75.784 mJ
P_average = 75.784 mJ / 600 s ≈ 126.3 µW
E_day = P_average × 24 h ≈ 3.03 mWh
Although the node spends 99.7% of its time in sleep, the transmission (only 0.5 s out of 600 s) accounts for almost 80% of the energy consumed in the cycle. Reducing the number of transmissions - for example by aggregating several readings before transmitting - can be more effective than further optimizing sleep or processing, which already contribute very little.
10Energy neutrality over time6 min
Comparing average energies is necessary, but not sufficient - the evolution of the storage must be analyzed over discrete intervals, step by step.
E_s[k+1] = E_s[k] + η_c·E_h[k] - E_c[k]/η_d - E_l[k], bounded by
0 ≤ E_s[k] ≤ E_max, where η_c and η_d are the charging and discharging
efficiencies, respectively.To verify a design, the following must be analyzed: the minimum level of stored energy reached during the simulation, the number of consecutive days without sufficient production, the energy lost when the storage is full (wasted harvesting), the periods when service must be reduced, and the time needed to fully recover after a significant discharge.
11Energy storage9 min
The storage element separates energy production from consumption in time. The choice is based on the total energy needed, the power peaks, the number of cycles and the environmental conditions.
| Characteristic | Battery | Supercapacitor |
|---|---|---|
| Energy density | high | low |
| Power density | moderate | high |
| Number of cycles | limited | very high |
| Voltage under discharge | relatively stable | drops continuously |
| Typical use | long-term storage | power peaks, frequent cycles |
E_B,usable = V_nom·C_B·DOD·η_d (DOD = allowed depth of
discharge). Supercapacitor: E_SC,usable = ½C(V_max² - V_min²).The system must operate for N_A = 5 days without harvesting, with E_consumption,day = 24 mWh, V_nom = 3.7 V, DOD = 0.8, η_d = 0.9. What is the minimum battery capacity?
See the solution
C_B,min = (N_A × E_consumption,day) / (V_nom × DOD × η_d) = (5 × 24 mWh) / (3.7 × 0.8 ×
0.9) ≈ 45 mAh
In practice a larger capacity is chosen, to account for battery aging, reduced capacity at low temperatures, self-discharge and the uncertainty of the energy model - 45 mAh is a theoretical minimum, not a final design target.
P_load = P_battery
+ P_supercapacitor. The supercapacitor reduces the peak current drawn from the battery,
lowering voltage drop, internal heating and the degradation caused by repeated pulses. The hybrid
architecture is justified when the load has low average power but demands short, repeated current
pulses - exactly the typical profile of an IoT node with periodic radio transmission.12Energy-autonomous sensor nodes8 min
Integrating all the concepts above into a real sensor node requires coordinating the energy source, the storage, the sensors and the communication strategy - none of them alone guarantees autonomy.
The typical architecture of an energy-autonomous node includes: the harvesting source and its converter, the PMIC with a cold-start function, the storage element (battery, supercapacitor or hybrid), the microcontroller with low-power modes, the sensors, and the communication interface (usually the most energy-costly part).
Adapting the service to the available energy can act on several parameters at once: the sampling period, the number of transmissions, the radio power, the measurement precision, the processing complexity and the number of active sensors. Choosing the right combination depends on which of these parameters affects service quality the least per unit of energy saved - an analysis specific to each application, not a universal recipe.
The hardware architecture, the measurement frequency, the communication strategy, the software algorithms, the storage capacity and the behavior explicitly defined for unfavorable energy conditions must all be adapted together - an energy converter attached to a node designed without regard for these constraints rarely produces real long-term autonomy.
13Frequent mistakes5 min
- "If the average harvested energy exceeds the average consumption, the system is certainly energy neutral." False - the average condition does not account for the distribution over time. A system can have a favorable annual average and still discharge completely during an unfavorable consecutive period. Analyze the evolution of E_s[k] step by step, for the most credible worst-case scenario, not just the average.
- "A 90% efficiency at each stage of the conversion chain means an efficient system." The overall efficiency is the product of the individual efficiencies - four stages at 90% each produce only about 66% overall efficiency, a loss easy to underestimate if the stages are viewed in isolation. Always calculate η_overall as the product of all the chain's stages, not just the most visible one.
- "The nominal power of a converter correctly describes the energy available in the application." No - the nominal power is measured under ideal lab conditions. The real usable energy depends on availability, orientation, aging and the application's concrete worst-case scenario. Size the design based on the estimated usable energy under the application's real conditions, not on the component's datasheet.
14Summary and glossary5 min
An Energy Harvesting system replaces the certainty of stored energy with a source that is variable and unpredictable in the short term - correct design requires analyzing the worst-case scenario, not just the average. The main sources (photovoltaic, thermoelectric, vibration, RF) have very different power and availability characteristics, and choosing or combining them must be guided by the application's real profile. The power management circuit (PMIC) must handle rectification, impedance matching (MPPT), cold start and, often, threshold-based operation. State-based consumption modeling frequently shows that a short but high-power event (radio transmission) dominates the total energy, even though the rest of the time is spent in a minimum-consumption mode. Storage sizing must be done for the worst-case scenario - consecutive days without sufficient harvesting - and hybrid battery+supercapacitor architectures combine the advantages of each technology.
15Self-check questions6 min
- What is the general energy chain of an Energy Harvesting system?
- Why is the condition "average harvested power exceeds average consumed power" not sufficient?
- What does the maximum power point (MPP) of a photovoltaic cell represent, and why is MPPT needed?
- What is cold start, and why is it a problem specific to Energy Harvesting systems?
- Why is the overall efficiency of the conversion chain the product, not the sum, of the individual efficiencies?
- How is the minimum battery capacity calculated for a reserve of N days without harvesting?
- What role does a supercapacitor play in a hybrid storage architecture?
16Where to go next2 min
The next lecture, the last in the series, moves to communication in networks of IoT devices - how the nodes studied in this lecture communicate with the rest of the system, which protocols are suited to low energy consumption, and how everything is integrated into a complete architecture.
The source sizing, the storage sizing and the consumption-profile modeling discussed here become real code and real measurements in Laboratory 02, where you analyze the consumption states and the autonomy of an ESP32-C6 node.