Executive Summary

Internet of Things (IoT) has demonstrated a clear market need for low-power, wide-area networks (LPWAN). For many business use cases, sensors must have battery-efficient connectivity to unlock the true potential of IoT. Unfortunately, industry wide fragmentation and lack of standards have caused massive confusion over different implementations of LPWAN. The purpose of this White Paper is to clarify key concepts of LPWAN, compare and contrast the different technologies, and provide resources for further exploration.

 This White Paper is intended for a wide audience. If you have some familiarity with wireless technology or electrical engineering, you may skip to the LPWAN comparison section and skim the appendices for underlying technology summaries. For those who would like to understand the general concepts, we recommend you read the document in whole sequentially.

LPWAN Overview

This section provides a high-level summary of LPWAN, and why it was developed. It’s important to understand what sets LPWAN apart from existing wireless technologies, and why other protocols do not fit the needs of LPWAN-enabled devices.

What is LPWAN?

The first thing to understand is that LPWAN (Low Powered Wide Area Network) is NOT a standard. It is a broad term encompassing various implementations and protocols both proprietary and open-source that share common characteristics as the name

• Low power: Operates on small, inexpensive batteries for 7 - 10 years
• Wide area: Has an operating range that is typically more than 2 km in urban settings

A physical limitation to achieve low power and wide range is small data size. Most LPWAN technologies can only send less than 1,000 bytes of data per day or less than 5,000 bits per second. These characteristics make LPWAN an excellent choice for the following classes of IoT applications: 

1. Dense locations: cities or big buildings for smart lighting, smart grid, and asset tracking
2. Long term

Simply put, LPWAN technology works well in situations where devices need to send small data over a wide area while maintaining battery life over many years. This distinguishes LPWAN from other wireless network protocols like Bluetooth, RFID, cellular M2M, and ZigBee, shown below with regards to bandwidth and range capability.

Shortcomings of Existing Networks

Cellular: Cellular networks suffer primarily from poor battery life and have gaps in coverage. Another difficulty is technology sunsetting: there are 30 million 2G endpoints in the US orphaned by sunsetting. Many of the IoT devices must remain on the network for 10 years. It doesn’t make economic sense if a cellular network is sunset and no longer supported.

Cellular LPWA (NB-IoT): Most notable aspects of cellular LPWA are still under development. As mentioned with cellular networks above, technology sunsetting is a huge concern. Cat-0 was touted as a long-term solution, only to fall victim to sunsetting. There’s active research into LTE-M, NB-IoT, EC-GSM, and 5G IoT, but none of them are cross-compatible and may not be suitable for long-term IoT solutions at the moment. Limited testing is being performed now, but a public-accessible service most likely won’t be available for 12-18 months.

Mesh Networks: Mesh networks like ZigBee are being used in IoT applications. In fact, many home automation systems deploy ZigBee, but ZigBee isn’t an ideal fit for LPWA applications. Mesh networks are only useful at medium distances and do not have the long-range capabilities of LPWAN technologies. More importantly, mesh networks are not battery efficient. Each node must constantly receive and repeat the neighboring RF signals. When sensors scale to the thousands, ZigBee does not fit the needs.

 Local RF: Bluetooth and NFC simply do not have the range to be useful for LPWAN applications.  

FAQ

Then what is LoRa, SigFox, Ingenu, Weightless-P, and nWave?

LoRa, SigFox, Ingenu, Weightless-P, and nWave are different implementations of the LPWAN technology. The difference in each implementation is discussed at length in the Key Player Comparison Section.

Does the requirement for long battery life force a star topology for all LPWAN applications?

While it is true that most LPWAN technologies use a star topology, you could theoretically use LoRa’s features, for instance, on a mesh network to achieve LPWAN (despite serious engineering challenges). Also, LoRaWAN (the MAC layer typically associated with LoRa implementations) uses a star on star topology, with Ingenu using a tree topology with an RPMA extender. Therefore, topology alone cannot be used to characterize LPWAN technologies.

If LPWAN protocols are so power efficient, why can’t we apply it to other protocols like WiFi to make it use less power?

The key thing to understand is that low power consumption does not mean great power efficiency. Most LPWAN protocols use small amounts of power, because they are either sending smaller amounts of data infrequently or sending them at a slower data rate. In fact, LPWAN protocols use more energy than 2G networks when transferring the same amount of data.

LPWAN Key Concepts

In order to properly compare the different LPWAN implementations, some of the concepts must be discussed beforehand to determine the appropriate application of each technology. If you are an electrical engineer or already familiar with information theory concepts, feel free to skip to the next section.

Link Budget

The main factor used to compare LPWAN technologies is the link budget. This is a key parameter in determining the range of the system. The link budget accounts for all the gains and losses in wireless signal transmission:

Received Power (dBm) = Transmitted Power (dBm) + Gains (dB) - Losses (dB)

Note the units of each variable. In signal transmission, decibel (dB) is used to represent the ratio between two power levels at a log scale:

dB = log(P1P2)

In contrast, dBm is an absolute power level measure in dB per 1 mW.

1 dBm = 10 log(P11mW)

Signals naturally attenuate over long distances due to propagation losses and signal attenuation through materials (see Appendix A for a detailed discussion). This means that to reach a wide area, the system needs alarge link budget to account for the losses.

Receiver/RF Sensitivity

Another important consideration with regards to range is RF sensitivity. RF sensitivity is the minimum magnitude of the input signal required to achieve the minimum error rate given a specific signal-to-noise ratio (SNR). In other words, more sensitive receivers can detect weaker signals since they can use less energy for demodulation.

For example, consider a device that can transmit +15 dBm and the signal experiences 140 dB of losses. The resulting signal at the receiver is then -125 dBm. Non-LPWAN receivers typically have sensitivities between -90 to -110 dBm. This means that it cannot detect the -125 dBm signal. LPWAN technologies have receiver sensitivities better than -130dBm, so they can hear this small signal despite heavy transmission losses. It is important to understand that a difference of 40 dBm means that LPWAN receivers are 10,000 more sensitive than non-LPWAN technologies due to the logarithmic scale. This becomes an important consideration for range discussions. So the greater the sensitivity, the longer the range.

 The graphic below summarizes the relationship between the link budget and RF sensitivity:

Modulation Rate & RF Sensitivity

From Information Theory, we see an inverse relationship between modulation rate and receiver sensitivity, which in turn affects the range. The Shannon-Hartley theorem models the maximum amount of error-free data over a noisy channel:

C=Blog2(1+SN)

where

• C: channel capacity (max data rate)
• B: bandwidth
• S: average received signal power
• N: average power of the noise

 The critical implication of the Shannon-Hartley theorem is the limit it imposes on energy per bit relative to the thermal noise spectral density (Eb/No). Consider a case where the transmission rate is equal to C (max data rate); then the average energy per bit is Eb/S where S is the average signal power. Total noise power is given by No B giving the following limit:

CB=log2(1+ EbNoCB)

EbNo=BC(2C/B-1)

EbNo=2-1

where η=C/B or spectral efficiency, the amount of data that can be transmitted per bandwidth.

 Since the  bandwidth is fixed,  designing a good modulation scheme becomes paramount to maximize η, which in turn maximizes the throughput of data. This means that the Shannon-Hartley theorem puts a lower limit of -1.6dB on the system. In other words, as η increases, Eb/No also must increase, so there is more power required per data bit.

 Returning back to RF sensitivity then, we see the role modulation rate plays:

Sensitivity = (-174 dBm + NF) + EbNo+10 log (Data Rate)

where

• -174dBm: thermal energy at room temperature per Hz
• NF: noise factor (2 - 5 dB)
• Eb/No: energy per data bit relative to thermal noise spectral density

 For example, if modulation rate is cut in half, we double Eb/No and increase the receiver sensitivity by half. But the tradeoff happens with battery life. Since more power is needed to push each bit, battery life inevitably suffers. Also, the data will remain in the air for a longer duration, meaning that there’s a higher likelihood of interference and collisions. SigFox is an example of using slow modulation to achieve longer range.

Noise Floor & Processing Gain

The Shannon-Hartley theorem places a significant power limitation on data capacity. One way to mitigate this is to use a narrowband signal with ultra narrow channel sizes. Since noise spreads throughout the spectrum, narrow bands give smaller noise levels. Another way is to have some processing or coding gain to build both a high coverage and high capacity network. The basic idea is to use spreading to maintain the bandwidth while allowing lower η to achieve higher receiver sensitivity. This is the fundamental idea behind chirp spread spectrum (CSS) modulations for LoRa-based implementations and code-division multiple access (CDMA) signals like Ingenu’s network.  

Interference & Performance

The theoretical performance for both a narrowband channel and a coded channel approach is roughly the same. However, since the Shannon-Hartley theorem applies only for a best-case scenario on a single-radio link (unlike the multiple access systems in many LPWAN implementations), there exists heavy debate as to which system performs better in the presence of interference.

 For narrowband channels, the signals are not affected by huge interference from their neighboring channels. At the same time, if interference occurs within the same channel, the signal will lose integrity fast. Coded channels collect widespread noise, which raises the noise floor. To compensate for the added noise, the system must send redundant data spread across many channels, resulting in very low spectrum efficiency.

Licensed vs. Unlicensed Channels

Major LPWAN technologies currently use a sub-GHz unlicensed ISM band (915 MHz in North America and 868 MHz in Europe). Although these technologies could theoretically work better in licensed bands due to less interference, most utilize unlicensed bands to avoid costly RF spectrum costs and retooling the MAC scheme. This affects coded signals more since, in unlicensed bands, you can get up to 26 MHz of spectrum, whereas in licensed bands you be allocated less than 1 MHz of spectrum.

 The biggest challenge right now is the fragmentation and lack of standard in using the sub-GHz spectrum. In Europe, the 868 MHz band is allowed and in North America the 915 MHz band is used. This means that there is no globally available sub-GHz band for LPWAN compared to Bluetooth and WiFi, which uses 2.4 GHz globally.  One of the distinguishing points of Ingenu’s RPMA technology is that it currently uses the unlicensed 2.4 GHz band and can operate internationally. The GSMA Mobile IoT initiative, however, is working on standardizing LPWAN in licensed spectrums. In June 2016, 3GPP decided to standardized NB-IoT, EC-GSM-IoT and LTE MTC Cat-M1, so this is a trend worth watching.

Orthogonality

Orthogonality is a feature of a coded channel where the two simultaneous signals are both detectable. Narrowband channels, like the FSK system used by Sigfox, can’t detect more than one signal at once; it will only detect the stronger signal if two signals use the same channel. Coded channels can detect multiple data streams using the same channel, which allows the system to use a wider band. This allows systems like LoRa/LoRaWAN to overcome the Shannon-Hartley limit.

Key LPWAN Players

As of December 2016, the following standards are under active development or deployment: SigFox, LoRaWAN, Weightless, Ingenu RPMA, and nWave. There are others such as LTE-Cat M, IEEE P802.11ah (low power WiFi), and Dash7 Alliance Protocol that are available, but not considered key LPWAN players based on adoption and initial traction. The following sections will briefly discuss each of the technologies and list pros and cons of each method.

 *** Disclaimer: The information presented here is based on publicly available research and should not be taken as a recommendation for using one service or another as it has not been empirically verified by Leverege yet. Given the pace of change within this sector, I recommend that any specific questions to be directed to the vendors mentioned below. Do, however, send us inaccuracies in the report if you spot them at team@leverege.com

SigFox

SigFox arguably has the most traction in the LPWAN space due to its successful marketing campaigns in Europe. It also boasts a great ecosystem of vendors including Texas Instruments, Silicon Labs, and Axom.

 It is a relatively simple full-stack technology whereby a binary phase shift keying (BPSK) modulation scheme is used to send data. It uses an ultra-narrowband of 100 Hz, sending very small data (12 bytes) very slowly (300 baud). SigFox is an example of using a slow modulation rate to achieve longer range. Due to this design choice, SigFox is an excellent choice for applications where the system only needs to send small, infrequent bursts of data. Possible applications include parking sensors, water meters, or smart garbage cans. However, it also has major drawbacks; downlink capabilities are severely limited and collisions become a huge issue. 

Pros:

• In deployment with a lot of traction
• Great relationship with vendors (TI, Silicon Labs, Axom)
• Power-efficient: no RX circuitry so they  consume less energy
• Great for simple monitoring, metering applications

‍

Cons:

• Not an open protocol, limited by SigFox networks
• Low Security: 16 bit encryption
• Limited use cases: unfit for cases where downlink communication is critical
• FCC regulation: SigFox transmission is too long for the limit set by the FCC under Part 15. So the architecture in the US is significantly different than the existing, tested ones in Europe.
• High levels of interference

LoRa-Based Standards

LoRaWAN and Link Labs’s Symphony Link both use LoRa as the physical layer and build their own MAC layer for their implementation of the LPWAN technology. In essence, LoRa refers to the chip manufactured by Semtech that uses chirped spread spectrum (CSS) technology to modulate the signal. For a detailed discussion of LoRa, refer to Appendix B.

LoRaWAN

LoRaWAN is an open-standard governed by the LoRa Alliance. However, it’s not truly open since the underlying chip to implement a full LoRaWAN stack is only available via Semtech. LoRa chips by themselves do not mitigate collisions very well. LoRaWAN is a multiple access protocol to handle this issue. It is laid out in a star-on-star topology to relay messages to the central server using gateways.

The functionality is similar to SigFox in that it is primarily for uplink-only applications with many end-points. Instead of using a narrowband, however, it spreads out information on different frequency channels and data rates using coded messages. These messages are less likely to collide and interfere with one another thereby increasing the capacity of the gateway.

LoRaWAN actually has three classes:

• Class A (Bi-directional end-devices): Each uplink transmission is followed by two short downlink receive windows. Class A devices use the lowest power and are useful for applications where downlink communication is not critical (similar to SigFox)
• Class B (Bi-directional end-devices with scheduled receive slots): In addition to the two downlink receive windows, Class B devices schedules a synchronized beacon to let the server know that the end-device is listening.
• Class C (Bi-directional end-devices with maximal receive slots): Class C devices have open receive windows that are only closed when transmitting. This uses a lot of power and renders it unfit for long battery-life applications.

Pros:

• Large, influential members including Cisco, IBM, Kerlink, Actility, and SK Telecom
• Better security: AES CCM (128-bit) encryption and authentication
• Flexible packet size defined by the user
• In deployment, most popular along with SigFox (over 100 commercial operators)

Cons:

• Not good for private/customer-deployed networks: having a central server helps mitigate collision issues
• Downlink capability is still limited
• Limited to Semtech-approved vendors
• ALOHA-type protocol makes validation/acknowledgment difficult; can have error rates over 50% in extreme cases

Symphony Link (Link Labs)

Link Labs is a LoRa Alliance member that built their proprietary MAC protocol on top of Semtech’s LoRa PHY. It adds the following features on top of the LoRaWAN protocols:

• Guaranteed message receipt
• Firmware upgrade over-the-air
• Removes duty cycle limit
• Repeater capability
• Dynamic range

In summary, the biggest difference is that it adds a synchronous layer to allow repeaters and acknowledgment messages. Also, before every transmission, the end device calculates the reverse link to the gateway and dynamically adjusts its parameters 

Pros:

• High sensitivity (same as LoRaWAN): -137 dBm
• Flexible frequency/no duty cycle limit: 150 MHz to 1 GHz (both unlicensed and licensed)
• Added features to LoRaWAN protocol such as the ability to operate without the network server

Cons:

• Requires Symphony Link software (added dependency)
• Smaller community of users

Ingenu (formerly On-Ramp Wireless)

Ingenu believes that using Random Phase Multiple Access (RPMA) is the best way to deliver a robust LPWAN solution. As a founding member of the IEEE 802.15.4k task group (which was dedicated to low-energy infrastructure monitoring), Ingenu put in a tremendous amount of effort in developing its technology stack, whereas SigFox and LoRaWAN groups have focused on faster time to market.

Due to its architecture, it has a superior uplink and downlink capacity than other models. It claims to have better doppler, scheduling, and interference robustness. It also operates in the 2.4 GHz spectrum, which is globally available (used for WiFi and Bluetooth). This means that there are no architecture changes per region like SigFox.

According to its internal studies, RPMA has a significantly higher link budget (177 vs. 149 and 157 for SigFox and LoRa respectively), meaning better coverage:

Pros:

• Good technology stack
• High coverage and robustness
• Gaining commercial traction despite late market entry

 Cons:

• Uses 2.4 GHz so more interference from WiFi and Bluetooth
• Propagation loss is increased at the higher frequency
• Structural penetration (buildings, walls, etc.) is not as good at the higher frequency
• Technology uses more processing power (may not fit the long-battery life criteria)

 Weightless

Weightless is the only truly open-standard that operates in sub-1 GHz unlicensed spectrum. There are three versions of Weightless that serve different purposes: 

• Weightless-W: leverages white space (unused local spectrum in licensed TV band)
• Weightless-N: unlicensed spectrum narrowband protocol born out of NWave’s technology
• Weightless-P: bidirectional protocol born out of M2COMM’s Platanus technology

 Weightless N and P are the more popular options since Weightless-W has a shorter battery life.

Weightless-N/NWave

NWave is very similar to SigFox in terms of functionality but boasts a better MAC layer implementation. It claims to use “advanced demodulation techniques” to allow its network to coexist with other radio technologies without additional noise.

Weightless-P

This standard uses FDMA+TDMA modulation in 12.5 kHz narrow band (greater than SigFox but less than LoRa). It also has an adaptive data rate, similar to Symphony Link (200 bps to 100 kbps). The sensitivity is quite high, -134 dBm at 625 bps and supports both PSK and GMSK modulation.

Resources

Comparison of LPWAN technologies are difficult since key parameters such as range and cost of deployment depend on external factors including density of urban environment, level of interference, and location of the receiver/transmitter. To make things more complicated, most of the available research has only been tested in Europe where the operating frequency is different than the deployments in the US.

 With these caveats in mind, here are some resources to dig deeper into each option:

• Waviot: NB-Fi vs. other LPWAN comparisons (note that WAVIoT supports NB-Fi)
• Ingenu: Making of RPMA ebook offers a deeper look into its technology
• Maarten Weyn: LoRa packet collision simulation - shows the level of interference experienced by LoRa networks

Appendix A: Sources of Signal Loss

To account for an accurate link budget, various sources of losses must be considered. The following list covers the most prominent sources of signal loss.

Propagation Losses

RF signals dissipate as they spread out. In free space, the dissipation grows as the square of the distance:

FSPL = (4d)2

where:

• d: distance from the transmitter (m)
• λ: signal wavelength (m)

Along the ground, dissipation grows quadratically with distance due to self-interference after the signal bounces off the ground:

PrPt(G4d)2(1-(1-j))2

= Pt(G4d)2(-4hthrd)2

= -PtG (ht hr)2d4

Attenuation Losses

Signals attenuate as they propagate through materials other than air such as building materials and interior walls. The losses can be significant depending on the material and the frequency of the signal. The NIST Electromagnetic Signal Attenuation in Construction Materials guide provides detailed information on signal propagation through various materials and is considered the go-to resource in the RF industry.

Losses through Connectors and Antennas:Non-ideal antennas have efficiency limits imposed by the radiation power factor:

16Abl3

where:

• Ab: cylindrical volume of the antenna
• l: radian length (1/2𝜋 wavelength) at operating frequency

‍

Appendix B: LoRa

LoRa is a proprietary PHY layer made by Semtech. Therefore, the underlying technology is not fully open. This section will analyze open parts of LoRa and include empirical analysis from other researchers and vendors.

Overview

LoRa is a modulation technique that uses chirp spread spectrum technology. LoRa chips generate a stable chirp using a frac-N phase lock loop (PLL), which is a linear variation of frequency over time. Due to this linearity, LoRa receivers are not affected by the Doppler effect and can be built inexpensively since extreme accuracy to account for frequency offsets are not necessary. LoRa receivers can achieve a sensitivity in the order of - 130 dBm even with cheap crystals.

 Compared to traditional frequency shift keying (FSK) and phase shift keying (PSK) modulation schemes, LoRa receivers have a higher out-of-channel selectivity and co-channel rejection of 90 dB and 20 dB respectively.  LoRa can also demodulate several orthogonal signals at the same frequency if they have different chirp rates (spreading factors with higher spreading factors denoting slower chirps).

Parameters

LoRa chirp rate is only dependent on the bandwidth. In fact, the chirp rate is equal to the bandwidth. Knowing that a LoRa symbol is encoded by 2SF chirps (where SF is the log2 spreading factor) that covers the entire frequency band, the relationship between the chirp rate and bandwidth brings several important consequences:

1. Spreading factor is inversely proportional to the frequency span of a chirp and directly proportional to the duration of a symbol.
2. Spreading factor does not affect the bit rate.
3. Bit rate at a spreading factor is proportional to the frequency bandwidth.

 LoRa also includes forward error correction code where the code rate is 4/(4+n) with n ∈ {1, 2, 3, 4}. So the useful bit rate is given by the following:

Bit Rate = SF BW2SFCode Rate

In general, receiver sensitivity increases with spreading factor and decreases with more bandwidth. So the theoretical receiver sensitivity per spreading factor and bandwidth are tabulated as following from the SX1276 datasheet:

Format

LoRa uses a specific structure to transmit physical frames:

1. Each message begins with a preamble where the up-chirps cover the entire frequency band and encodes a sync word. The sync word differentiates the LoRa network from others in the same frequency band.
2. The optional header provides the size of the payload, code rate, and the presence of payload CRC.
3. Payload and the optional CRC follows the header.

‍

Appendix C: Symphony Link Protocol in 900 MHz

1. Scan the band and create an interference profile
2. Gateway selects a 500 kHz channel for its downlink
3. Systems begins transmitting every 2 seconds
4. Encrypted with the network ID - makes network private with application token
5. uplink/downlink time boundary (LoRA is half-duplex technology, so important to prevent up/down collisions)
6. Uplink channel frequencies