For IoT applications, the main characteristics of the physical layer that need to be considered are modulation, data rate, transmission mode, and channel encoding.
Modulation. The nature of IoT applications, some involve infrequent data transmission that need low-cost low-complexity devices, preclude the use of high-order modulation or advanced channel coding like trellis-coded modulation. Unless mandatory, due to a harsh radio environment with narrowband interferers or regulatory constraints, spread spectrum, e.g., Direct Sequence Spread Spectrum (DSSS), is to be avoided as it increases the channel bandwidth, requiring a more costly and power-consuming RF frontend, with no data rate improvement. Allowing non-coherent demodulation relaxes the constraint on the device complexity, so (Gaussian) Frequency Shift Keying ((G)FSK) is a proven and suitable choice, similarly as in Bluetooth radio. It is considered that the most sensible choice upon availability would be Gaussian Minimum Shift Keying (GMSK), as the modulation index of ½ allows for lower complexity, or better sensitivity at a given complexity. When available bandwidth is restricted, GFSK with lower modulation index is still appropriate, with the next best being 1/3 as it still allows for near-optimal demodulation at reasonable complexity.
Data rate. IoT applications need to mix very low data rate requirements, e.g., a sensor or an actuator with limited data size either uplink or downlink, with more demanding requirements, e.g., a 6-inch 3-color ePaper display in a home that updates the daily weather forecast or the shopping list, easily amounting to more than 196 kB worth of data. Yet even for small data amounts, a carefully chosen higher data rate actually improves power-consumption thanks to shorter transmission time and reduced probability of collision. Similar reasoning is applied to Bluetooth Low Energy, a.k.a., BLE or Bluetooth Smart, formerly Nokia’s WiBree, which uses 1 Mbps with much lower data throughput. The latter is aimed at proximity communication and its high gross data rate of 1 Mbps sacrifices the range considerably. Even when operating at sub-GHz frequencies, which offer better range than 2.4 GHz for a given transmit power, the 1 Mbps is considered to be the absolute upper limit. On the higher end, the transceiver complexity and power increase do not improve the actual useable throughput, as the overhead of packet acknowledgement and packet processing time become the bottleneck.
On the lower end, data rates below 40 kbps are actually impractical, as it would rule out using standard off-the-shelf 20 parts per million (ppm) crystals. Indeed, the frequency accuracy of these crystals is not sufficient: 20 ppm translates into a 18 kHz frequency error when operating in sub-GHz bands, while it is 48 kHz when operating at 2.4GHz. A narrow channel requires an accurate crystal like temperature-compensated TCXO on both ends, including the client, which is more costly, power-consuming, and bulky .The optimal baseline gross data rate is considered to be 500 kbps. Depending on the scale of the network, e.g., home, building, district, or city, the applications, and the number of devices, we expect different trade-offs with actual deployments ranging from 100 kbps to 500 kbps.
Transmission mode. Full duplex communication is challenging, as it requires good isolation and does not allow for resource sharing between transmit and receive. Full duplex also typically involves different frequencies for downlink and uplink. Since the radio resource is a scarce resource, half-duplex is therefore selected, preferably on the same radio channel.
Channel coding. There is the potential for improving link quality and performance with a limited complexity increase by using (adaptive) channel coding together with Automatic Repeat-Request (ARQ) retry mechanism. As of today, this is considered optional due to complexity-cost-performance trade-offs achieved with current technologies. However, provisions have to be made for future implementation. As of today, flexible packet length is considered a sufficient means of adapting to the link quality variations.
For IoT applications, the main characteristics of the media access layer control (MAC) that need to be considered are multiple access, synchronization, and network topology.
Multiple Access. Looking back at decades of successful cellular system deployment, one can safely conclude that TDMA is a good fit for the IoT. TDMA is suited for low-power operation with a decent number of devices, as it allows for optimal scheduling of inactive periods. Hence, TDMA is selected for multiple access in the MAC layer.
Synchronization. In IoT applications, there are potentially a very large number of power-sensitive devices with moderate throughput requirements. In such a configuration, it is essential to maintain a reasonably consistent time base across the entire network and potentially across different networks. Given that throughput is not the most critical requirement, it is suitable to follow a beacon-enabled approach, with a flexible beacon period to accommodate different types of services.
Network topology. Mobile networks using a cellular topology have efficiently been servicing a large number of devices with a high level of security and reliability, e.g., 5,000+ per base station for LTE in urban areas. This typology is based on a star topology in each cell, while the cells are connected in a hierarchical tree in the network backhaul. This approach is regarded suitable for the IoT and is therefore selected.
The network layer (NWK) and the interface to applications are less fundamental as far as power-efficiency and reliability is concerned. In addition, there is more variation in the field of IoT applications. Nevertheless, it is widely acknowledged that IoT applications need to support the Internet Protocol (IP), whether it is IPv4 or IPv6. In addition, the User Datagram Protocol (UDP) and Constrained Application Protocol (CoAP) could provide the relevant trade-off between flexibility and implementation-complexity on resource-constrained devices.
Furthermore, the IoT will represent an immense security challenge, and it is likely that state-of-the-art security features will become necessary. As of today, we can assume 128 bits Advanced Encryption Standard (AES) for encryption and Diffie-Hellman (DH), or the Elliptic Curve Diffie-Hellman (ECDH) variants, can become the baseline for securing communication.