For the PDF version of this article, click here.

The rapid development of on-chip radio technology, with the ongoing evolution of miniature sensor technology, MEMS and the like, are producing unusual capabilities. This technology enables the creation of autonomous, integrated radio-sensor modules with powerful sensing and communications features, operating within large wireless networks. These wireless modules (nodes) operate independently, sensing the environment and forwarding the collected information across a wide area, economically and efficiently.

So it is no surprise that potential applications are almost unbounded. They range from wireless devices on factory floors to security and crisis management, as well as medical applications. Because such devices occupy small physical volumes and consume so little power, they transmit low-power digital data to a base station (hub) via the network where the information can be immediately analyzed, processed and used.

In addition, the application of this technology for medical and homeland security presents a major evolutionary step in the welfare and security of our society and a unique business opportunity supported by the U.S. government.

We shall define miniature wireless radio-sensors as integrated transmitters or transceivers that occupy less then one cubic inch, with sensors, antenna, controller and battery. The cost target in volume quantities was less than $50-100 per node in 2004 and is expected to decline to less than $25 by 2008. In fact, many of the platform systems are already smaller and cheaper, depending on sensor type, range and power.

An example of such a product is an invivo implant capsule that monitors blood glucose at regular intervals. In medical applications such as this, medication and/or an alarm can be delivered immediately and efficiently. Such devices, once they mature, may become the “holy grail” for people suffering from chronic diseases. In a different application, a seismic sensor network can monitor seismic and earthquake activity over a large geographic area and deliver alarm signals to authorities and population.

Regardless of whether the data are medical, seismic, industrial or otherwise, they can be recorded and forwarded via RF, efficiently and economically. “Sensor-Net” technology is about to explode and become ubiquitous in almost every facet of our lives.

A miniature wireless video (camera) platform is shown in Figure 1. It occupies approximately one cubic inch and is autonomous, with a battery, radio, camera (color imager plus lens), controller, antenna and temperature sensors — along with the circuitry that performs video processing and handles networking protocol.

The technology of on-chip radio has evolved rapidly in the last few years. Companies such as Micrel, Atmel, Chipcon, Nordic, RFMD, Maxim, Microchip, and Melexis have introduced a number of standard devices that can transmit information at low to moderate data rates.

Some devices supply only the RF capability. Of special interest are radios such as the Microchip rfPIC12xx and the Chipcon CC10xx that also offer a modem, as well as a processor/controller and even a networking capability. Such devices have resident software and require few external components. Most of these devices employ 0.18 micron or 0.25 micron, CMOS or BiCMOS technology.

Obviously, simpler devices have many applications as well, such as Micrel's MICRFxxx, which was designed to have a small footprint and low cost and, therefore, may require an external programmer.

A generalized block diagram of a radio transmitter is depicted in Figure 2. It employs a phase-locked loop (PLL) derived from a crystal-controlled oscillator (crystal footprint sizes can be as small as 3.2 mm × 2.5 mm and are continuing to shrink) to generate RF for both transmitting and receiving. Generally, we find that transmitters comprise a crystal oscillator — which governs the RF accuracy and can also serve as the clock to provide timing for the controller — a phase-locked oscillator (PLO), a VCO and a modulator, followed by a power amplifier (PA) to provide power control. The power amplifier usually includes circuitry to modulate the power output and to match the antenna.

Most standard devices in the market use amplitude shift keying (ASK) or frequency shift keying (FSK) or both. Data rates range from 10 kbps to 1 Mbps. Both modulation techniques realize the benefits of saving power using constant amplitude transmission.

Let us first examine the transmitter. In the example shown in Figure 2, the VCO, which requires an external inductor (pins 10, 11), drives the internal phase detector and charge pump via a prescaler. The other input to the phase detector is from the crystal oscillator (crystal terminals are 15, 16). The crystal oscillator requires external capacitors because on-board capacitors consume too much die real estate. The VCO output is connected to the PA. The PLL loop filter is set externally at pin 12. Because sensor-net systems use packet switching communications protocol, information is transmitted in bursts. Since they are battery-operated devices, power is always at a premium. It is advisable to turn the whole transmitter off, including the crystal oscillator, between bursts. Returning the transmitter to the active state (on) requires a wait time in which the crystal oscillator is restored to stable operation. The estimated time (T_st) required to reach stable operation is approximated by Equation 1:

T_st ≈ 2Q/ƒ_x

where Q is the quality factor of the crystal and ƒ_x is the frequency of operation.

Typical 20 MHz crystal oscillator wake-up times, for Q values ranging from 10K to 500K, are plotted in Figure 3.

For a 20 MHz crystal with a Q of approximately 100,000 — a typical Q for miniature crystals at this operating frequency — the start-up time is about 10 milliseconds. This time is consumed (wasted), unless special circuitry is added to speed up the onset of stable oscillation. The start-up interval can be significant. Consider an overall burst at a data rate of 20 kbps: If the burst is 500 bits, including a synchronization sequence and a hand-shake protocol, the total duration is approximately 25 milliseconds. Therefore, the oscillator start-up interval adds a 40% overhead. When used, speed-up circuitry applies a burst of current via an AGC amplifier, increasing the oscillator gain momentarily until the output stabilizes.

Voltage-controlled oscillators (VCO) present a challenge in miniature radio design. In cases where FM-FSK or PSK modulation is employed, phase noise must be kept under control. Ring oscillators exhibit poor phase noise performance due to their low Q, which is on the order of 1. Hence, they exhibit a typical phase noise of -85 dBc/Hz at a 100 kHz offset when running at 900 MHz. LC oscillators, which are used predominately in such radios, will deliver much better phase noise performance, an improvement over a ring oscillator on the order of 15 dB to 20 dB.

LC oscillators face certain noise limitations caused by flicker noise (CMOS oscillators can exhibit a corner frequency of up to 0.25 MHz), circuit leakage (digital signals leaking into the radio chip), and by resonator Q. Printed (on die) inductors on CMOS exhibit low Q. Therefore, many applications require external inductors. External inductors also allow designing chips to cover a wide frequency range, usually 300 MHz to 1000 MHz. External chip inductors exhibit a high Q.

Most on-chip designs use a differential VCO structure, as shown in Figure 4, with an internal varactor. Above 1.5 GHz, bond wires can be used for discrete inductances as well as for printed coils.

In general, the design calls for two inductors. A single inductor usage, given the objective of minimizing the number of external components, is connected between the two outputs with dc supplied to one or both leads via a resistor or active stages.

In certain applications, the VCO inductor can also be used as the antenna, but with limited power delivery capability. This is a classical approach to generating a power-VCO structure, but with limited modulation options and power control range. In the case of ASK, phase noise is obviously not an issue.

Most applications handling complex accurate waveforms, such as GFSK, use quadrature modulators because they will support any modulation scheme and provide unmatched accuracy. Quadrature modulators require the generation of quadrature RF signals using mostly complex RC networks, or a divide by 2. Most integrated quad modulators achieve accuracy of 1° to 3°, with carrier and unwanted sideband attenuated by 30 dBc to 35 dBc. A block diagram of a quadrature modulator is shown in Figure 5.

Baseband I and Q signals can be stored in memory and require D/A converters. The D/As usually require no more than 5 to 6 data bits, a total of less than 64 bytes, for equalization or spectral control. The quadrature modulator will, therefore, include the ROMs, the DACs, the phase-shifting networks, the Gilbert cells and an RF amplifier. This configuration ensures the ultimate in modulation accuracy, but at a significant cost in current consumption.

Simplified chips can use on/off for ASK. For FSK, they can modulate the VCO directly. In such cases, the VCO will have two varactor bridges: one for PLL lock (wideband) and one for modulation.

RF power control can be achieved by analog or digital modulation and seldom requires a range of more than 20 dB to 30 dB. Algorithms for overall power control within the network are governed by the radio's application with regard to the type of information and environment — such as a single device, a part of network and propagation conditions.

Device power can be controlled in two ways. In the case of direct modulation of the VCO, where the resonator also serves as the antenna, the VCO current can be controlled over the dynamic range and can achieve up to 20 dB — perhaps the current will change from 0.1 mA to 1.5 mA before the device starts to be saturated. Alternatively, in cases using a PA, the PA output power can be controlled by the current applied to it or can be followed by an attenuator. However, the latter wastes power. Of course, on/off control of the output stage is used in ASK modulation schemes.

In the case of a network, transceivers must be used. Figure 6 shows the block diagram of an integrated transceiver from Chipcon (Norway)^[1]. The radio chip includes the modem (modulator, demodulator) in addition to the complete transceiver. Some of these devices also include a processor (8051 type) or a networking stack. These products usually operate in the ISM bands and cover 300 MHz to 1000 MHz and 2500 MHz (ISM).

The Chipcon transceiver shown uses FM (FSK, GFSK) and modulates the PLL circuitry directly (no quadrature modulator). It comprises a built-in synthesizer, a VCO (external inductor), PA, T/R switch, crystal oscillator, a complete receiver chain and the modem — including demodulation — which it processes either at baseband or at a low IF. Over the 900 MHz to 2500 MHz frequency range (separate devices), the current requirement in transmit is 10 mA to 25 mA and 10 mA to 20 mA in the receive mode. The crystal oscillator circuit consumes approximately 250 µA.

To summarize the properties of the Chipcon-type integrated transceivers, the output power is 0 dBm to 10 dBm, output power control range is approximately 30 dB, and receiver sensitivity is - 90 dBm for a 250 kbps data rate. The noise figure is approximately 15 dB. The transceiver has a built-in T/R switch. The synthesizer switching time, if frequency hop is applied, is about 200 microseconds, and the PLL loop bandwidth is in the vicinity of 30 kHz.

Much has been written about the future of MEM sensors^[2]. Many have predicted that MEM sensors, as part of nanotechnology, will become part of the next technological revolution. There is no question that nanotechnology sensors for medical and biotechnology, as well as other sensing, will have a significant impact on our lives — whether it is in genetics, disease detection, homeland security, factories or food processing.

The evolution of MEMS is ongoing. Because MEMs can be manufactured in miniature sizes and at low cost, they are attractive for scientific, industrial and medical applications. To date, devices such as valves, levers, plates, chemical sensors, RF components and switches have been implemented as MEMs. They are fabricated using semiconductor processes and by bulk micromachining. Though some are already mature, most are still in their infancy.

MEM microphones appeared in the market last year, and miniature microphones have been in the marketplace for some time. Typically they have a footprint of 3 mm × 3 mm and require 2.5 V at 100 µA. Many other MEM sensors are already in use, including switches, reed relays, temperature and pressure sensors, accelerometers, and motion detectors — as well as various chemical detection devices. A number of miniature accelerometers and pressure gauges have been developed for applications in the automotive industry — for air bag control and automated tire pressure measurements.

Though today's seismic sensor devices are relatively large, efforts are under way to decrease their size and to achieve better control by deploying larger numbers, via distributed networks, at much greater densities per unit area.

Great progress has been made for fire and fume detection, for which miniature sensors already exist. Automotive and chemical MEMs can be used as pH meters. Dedicated sensors have also been developed as body-chemistry sensors, such as blood-glucose sensors. They show promise for significant medical advancements. Other sensors include imagers, from visible light to various IR and UV meters. Imaging sensors, which are migrating from CCD to CMOS (CMOS has lower sensitivity but allows system-on-chip integration and lower power), are already available in great variety, boosted by the cell phone industry. (For examples, look up components marketed by Micron Technology and Omnivision.) Some imagers already perform significant video processing, including compression. JPEG is already available on board some cameras (look up Transchip, Agilent).

Figure 7 displays chemical nonosensors — or “smart dust,” as they are sometimes referred to. They change colors and pattern in the presence of certain chemicals and are therefore useful as detectors in medical, industrial and homeland security applications.

MEM RF components, such as filters, sources, resonators, and switches, are under development[4]. Other miniature sensors, MEM and otherwise, are available from multiple sources.

The transmission of data by radio sensors is especially difficult because of the overall requirements for low power and the limitations imposed on the antenna because of its small dimensions. Such antennas, which typically have an overall geometry that is less than 10% of wavelength, exhibit low transmission efficiency and require tuning. Though various types of miniature antennas and radiators have been developed, the literature on them is limited. Radiators usually take the form of coils or printed and wire loops^[1].

To summarize the requisite antenna properties, we can assume that the antenna can be represented by a parallel resonance circuit with the inductor L serving as the radiating element, as shown in Figure 8.

Capacitor C resonates the antenna, and the resistor R represents the distributed resistance of the inductor. If the radio operates at an angular frequency (ω_o) then the Q is given by:

Q = R/ω_oL (Eq. 2)

For an N-turn loop air coil, with diameter D, and total length l, all in mm, the inductance is given by:

L_nH ≈ N² × D²/(D + l) (Eq. 3)

For N = 3, the inductance as a function of diameter (in mm) of air coil is plotted in Figure 9.

The coil can therefore be modeled as a pure inductor in series (or parallel) with a resistor, due to finite Q, and a radiation resistance, which is usually of a small value in such an antenna. The radiation resistance R_rad, governs the amount of power the antenna will transmit, as given by I²R_rad, where I is the current through the coil.

R_rad is approximated by:

R_rad ≈ 20 × (2π/λ)⁴ × (NA)² (Eq. 4)

where: A is the area of the coil, and N is the number of turns. The radiation resistance for various coil radii is depicted in Figure 10.

For a radius r = 3 mm, the radiation resistance of the coil at 900 MHz turns out to be approximately 0.02 Ω.

We can now demonstrate the efficiency of such a system antenna. We assume that our antenna system has a loaded Q of 30. Our coil has a value of approximately 30 nH, which has an impedance of j130 × at 900 MHz.

Let us assume that we develop 3 V across the antenna. The resistor R has a value of 130 × 30 = 3900 Ω, and the current though the circuit, assuming resonance, is governed by:

I = V/R = 3/3900 = 0.77 mA. (Eq. 5)

The power dissipated in the resonator is given by:

I²R = (0.77)² × 3900 ≈ 2.3 mW = 3.6 dBm. (Eq. 6)

The current through the coil is Q times higher than in the resistor, or approximately I_r = 23 mA. The radiated power P_rad is I_r² × R_rad ≈ -20 dBm. Obviously, the radiation of this tiny antenna is minute, given that the radiation efficiency is less than 0.5%.

As antenna geometry grows relative to wavelength, the radiation resistance will improve, of course. In many applications, however, especially in the 450 MHz and 900 MHz range, node size prohibits even the use of a ¼-wavelength dipole. Ceramic materials help a bit as their dielectric properties allow certain geometry shrinkage.

The radiation pattern for such tiny radiators is much like an infinitesimal dipole, as shown in Figure 11.

Here are a few conclusions relating to the antenna:

• The antenna must be at, or close to, resonance, otherwise its efficiency decreases rapidly due to the lower current through the inductor.

• Radiator efficiency improves with the square of the antenna area.

• Radiator efficiency improves with antenna volume.

• Efficiency of such an antenna should be expected to be less than 1% to 5%, depending on the details of the design and the antenna's size relative to a wavelength.

Note that efficiency was calculated per resonator dissipated power — and not based on VCO or PA power.

Resonating the antenna can be achieved in several ways:

• When the coil is part of the VCO resonator, then the antenna is always in resonance. However, output power is limited (impedance matching issues).

• If the VCO is followed by a PA, various resonating techniques and algorithms must be developed — such as hill climber, checking of maximum voltage on resonator, or various phasing techniques.

Of course, certain antenna impedances can be transformed or matched to enable lowering of the impedance and thereby developing higher power levels, given the low voltages available in tiny silicon platforms. See the design example in Reference 3 for a printed loop, as well as Figure 12.

The battery is a major challenge in wireless sensor technology as it is usually the component that governs the length of time the node will function without battery replacement. Power management algorithms, therefore, become essential. Most customers would like to purchase sensor networks that operate years without battery replacement. As it turns out, in many of the applications we are discussing, the battery consumes up to 60% to 75% of the sensor module's (node) total volume.

Standard battery voltages are usually multiples of 1.5 V. A standard-alkaline, 1.5 V button battery can supply approximately 40 mAH to 50 mAH of energy. AA batteries provide 1000 mAH. Though batteries are readily available from many companies, their characteristics must be studied carefully. For instance, many batteries have been designed for ultralow-current applications, on the order of microamperes to power wristwatches.

Li-ion batteries have a larger capacity. A 1-inch long, 8-mm diameter lithium battery delivers 3 V with close to a 1000 mAH capacity. See, for example, the CR1/2 from Varta or the CR2 (popular digital camera battery) available from many other battery suppliers. While the li-ion based cell is popular due to its low weight and relatively large capacity, other electrolyte materials are being used as well — such as zinc-air, which has excellent capacity but is not rechargeable. Since data transmission in the network is characterized by low duty cycles, batteries can last a long time, depending on data type, networking and power management algorithms.

The sensor platforms that we have been describing will usually operate within a network. Networking miniature sensing platforms enable connectivity across much greater distances via network interconnection, and by finding optimal paths from point-to-point via a distributed network. Fortunately, the subject of networking has been assisted by large research programs and investments. Depending on the application, mesh- or star-type networks are used, and certain standards are starting to emerge, with data rates ranging from 10 kbps to 1 Mbps^[1,4].

Zigbee supports rates of 20 kbps to 250 kbps, enabling the operation of large networks and providing fast access to the network that is nominally three milliseconds — or 10 times faster than Bluetooth. Zigbee supports star or mesh (peer-to-peer) networking, 16-bit to 64-bit addressing, carrier-sense collision avoidance, and link acknowledgment. Most networks, which are mostly ad hoc at this point, can use up to 83 channels in the 2400 MHz range and 20 in the 920 MHz range (Zigbee supports less). Modulation is mainly ASK or FSK. QPSK has lately been introduced with DSSS modulation. Zigbee, which seems to be emerging as a de facto standard, was designed for large networks — up to 64K nodes (with 16 bits addressing) — for low-power, short-message, short-range applications.

The operating system of these networks is also unique. It is known as Tiny-OS, with its origins stemming from work at UC Berkeley. Various versions have been custom-designed but without an industry standard as of yet. This is why most of the existing operating systems have been designed ad hoc. Actually, many systems will most likely use a certain level of ad hoc protocols until consolidated into a standard. The transmission of large files, including video, over such networks definitely requires ad hoc as of now. Frequency hopping is used mostly for networking and for contending in multipath scenarios. Most sensor-net systems use packet switching.

The limitations of energy source, low RF output power and small antennas all contribute to the short range of these sensor modules. What's more, the range depends on propagation conditions. Fortunately, the adverse effects of these constraints can be mitigated by several techniques. Among them are frequency hop, which is already widely used in the 802.11 standards and in Bluetooth — and also antenna space diversity. The latter was pioneered over a half a century ago in HF communications.

Figure 13 is a chart of path loss at 1.6 GHz, assuming both antennas deliver 0 dBi gain. Transmitted power is 0 dBm, and receiver sensitivity is -100 dBm.

Most wireless sensor-net designs, if used in unattended ground applications, assume a range of approximately 30 meters. Such systems use only the radio chip for their RF (with no additional circuitry) and have, therefore, a total dynamic range of approximately 9 dB (0 dBm transmission and -90 dBm sensitivity). Obviously, for connectivity, large quantities will have to be distributed in this application.

In conclusion, though radio-sensor technology and products are just beginning to emerge, they have already demonstrated their potential to improve our productivity, security, and ability to study and better control our environment. Thanks to the advances in silicon-device technology, MEM sensors, nanotechnology and networking; wireless sensors show promise for significant challenges and manifold opportunities in the future. A rapid proliferation of the technology in government, industrial and home improvement applications is already in progress. So it is no surprise that the market for radio-sensor technology devices is expected to reach $20 billion by 2010.

1. Chipcon Corp., “An IEEE 802.15.4 and Zigbee-ready 2.4GHz RF Transceiver”

2. Hector De Los Santos, RF MEMs Circuit Design, Aetrech 2002.

3. Nordic Corp. Application Note AN400-03.

4. Zigbee, IEEE 802.14.5

B. G. Goldberg, PLL seminar notes, CEI, 2003.

“Wireless Sensor Networks, From Dust to Reality,” by ON WORLD, July 2003.

Micron Technology data sheet of VGA imager, Part No. MT9V112.

Sailor, Trogler, Sohn, Calhoun, UCSD, “Detection of TNT and picric acid on surfaces using photoluminiscent polysiloles.”

Bar-Giora Goldberg has acquired extensive experience in the design of radio communications systems. He received his Bachelor and Master of Science degrees from Technion-Israel Institute of Technology in Haifa, Israel. Before emigrating to the U.S., he established himself in the Israeli communications, radar, and spread spectrum markets. Goldberg holds six U.S. patents and four pending patents. He has published numerous articles in professional magazines and has written the book entitled Digital Techniques in Frequency Synthesis (McGraw-Hill 1996).