In the quest to build a world of "energy-autonomous" IoT, researchers have long relied on a convenient mathematical fiction: the linear energy harvesting model. This model assumes that the power a device harvests is directly proportional to the radio frequency (RF) power it receives. However, as Bruno Clerckx, Rui Zhang, and colleagues argue in Wireless Transmission of Information and Power—Part II, this idealized assumption creates a significant gap between theoretical optimization and physical reality.
The emerging consensus across the field is that we must move toward non-linear energy harvesting models to design systems that actually work on real hardware.
The Linear Lie and the Rectifier Reality
The heart of any wireless power receiver is the rectifier—a circuit that converts oscillating RF waves into steady DC electricity. In a linear model, if you double the input power, you double the harvested energy. But physical diodes and circuits don't behave this way.
Real-world energy harvesting exhibits two critical non-linear behaviors:
- Sensitivity Thresholds: If the incoming signal is too weak, the diodes won't turn on at all. The harvested power is zero, not a small fraction of the input.
- Saturation Effects: Once the input power reaches a certain level, the circuit hits a physical limit. Adding more RF power doesn't result in more DC power; the efficiency drops as the hardware "saturates."
Yi Fang 0005 and Yiwei Tao demonstrate in Design of a Reconfigurable Intelligent Surface-Assisted FM-DCSK-SWIPT Scheme with Non-linear Energy Harvesting Model that ignoring these effects leads to "overly optimistic" performance estimates. When engineers optimize for a linear world, they often allocate resources to users who are actually in the "saturation zone," wasting power that could have been better used elsewhere.
Why the Model Mismatch Matters
The consequences of using the wrong model aren't just academic; they lead to system failures. Ming-Min Zhao and Yunlong Cai show in Robust Transceiver Design for MISO Interference Channel With Energy Harvesting that when the "map" of the wireless channel is fuzzy, linear models fail to guarantee that users receive enough energy to stay powered.
Similarly, in low-latency applications using short codewords, Behrooz Makki and Michele Zorzi point out in Wireless-powered relaying with finite block-length codes that throughput predictions become inaccurate under varying input power levels if non-linearities are ignored. The "optimal" codeword length or power-splitting ratio changes entirely once the hardware's saturation point is factored in.
Optimizing for the Real World
To fix this, researchers are integrating non-linear models—often based on sigmoid functions or piecewise linear approximations—into every layer of the network stack.
Aerial and Underground Power
In Completion Time Minimization in Wireless-Powered UAV-Assisted Data Collection System, Zhen Wang 0035 and Guopeng Zhang use non-linear models to plan UAV flight paths. By accounting for saturation, they prevent the UAV from hovering too close to a sensor and "over-charging" it uselessly, instead optimizing the trajectory to cover more ground. This logic extends to underground sensors, where Kaiqiang Lin and Yijie Mao have developed time-allocation strategies that respect the non-linear efficiency of soil-embedded hardware.
Intelligent Surfaces and Moving Antennas
New technologies like Reconfigurable Intelligent Surfaces (RIS) and Pinching Antenna Systems (PAS) provide even more control over the power-data trade-off. HongGuang Sun and Jiangbin Lyu show in IRS-Assisted RF-powered IoT Networks that while RIS can provide a massive boost to energy harvesting, the gains are capped by hardware saturation.
Meanwhile, Kunrui Cao and George K. Karagiannidis explore PAS in Performance Analysis of Wireless-Powered Pinching Antenna Systems, where antennas physically move to "follow" users. Without a non-linear model, these systems would overestimate their own reliability, as they wouldn't account for the sensitivity thresholds that define whether a moving user is actually "in range" for power.
Beyond RF: The Non-linear Energy Profile
The shift toward non-linear modeling is even reaching the software layer. In DiSCo: Device-Server Collaborative LLM-based Text Streaming Services, Ting Sun and Fan Lai note that even for on-device AI, linear models based on operation counts fail to predict battery drain. Factors like thermal management and hardware-specific optimizations create non-linear energy profiles that mirror the complexities found in RF harvesting.
By embracing these non-linearities, the research community is moving away from mathematical elegance and toward practical, deployable systems that understand the limits of the hardware they inhabit.