Tutorial: Part I, Part II

Information, Modulation, and Multiplexing
A periodic signal that persists indefinitely, without changing its amplitude, frequency, or Phase--a continuous wave (CW) signal--carries no information other than the fact that it is present. In order to convey data, a signal needs to change. We normally think of this change as a relatively slowly changing variation--modulation--imposed on the periodic signal, for example:

The function m(t) is said to contain the baseband information, and the relatively high frequency cosine function is the carrier. When the function m(t) is another sine or cosine (presumably of much lower frequency), we can make use of trigonometric identities (see Appendix 2) to rewrite the signal in a revealing fashion:

A sinusoidal modulation splits the carrier wave into two signals called sidebands, one above and one below the carrier, each displaced by the modulating frequency (Figure 5). While a continuous sinusoidal modulation is hardly more interesting or useful than a CW signal, this result suggests that when a signal is modulated, the resulting frequency spectrum becomes wider.

Signals of interest for RFID are generally digitally modulated. A digitally modulated signal is a stream of distinct symbols. A simple example with substantial relevance for RFID is on-off keying (OOK). The signal power is kept large (m = 1) to indicate a binary 1 and small or zero (m = 0) to represent a binary '0.' An example is shown in Figure 3.6. In OOK, each symbol is a period of fixed duration in which the signal power is either high or low. Each OOK symbol represents one binary bit, though other types of symbols can convey more than one bit each. Any circuit that can change the output power, such as a simple switch, can be used to create an OOK signal, and any circuit that can detect power levels can demodulate (extract the data from) the signal. For example, a diode--an electrical component that passes electrical current only in one direction and blocks current flow in the opposite direction--can rectify a high-frequency signal, turning it into pulses of DC. These pulses can be smoothed with a storage capacitor to produce an output signal that looks very much like the baseband signal m(t) (see Figure 17 in Chapter 2). If the diode responds rapidly, it can be used at very high frequencies. Modern diodes can operate up to over 1 GHz, allowing passive RFID tags to demodulate a reader signal using only a diode and capacitor.

Figure 5; A Sinusoidally Modulated Carrier Wave and Corresponding Frequency Spectrum; fc is the Carrier Frequency.

Unmodified OOK is admirably simple and seems promising as a method of modulating a reader signal. However, there is a problem with OOK for passive RFID. As we noted in Chapter 2, a passive RFID tag depends on power obtained from the reader to run its circuitry.

Figure 6. On-Off Keyed Signal

If that power is interrupted, the tag cannot operate. However, imagine the case of an OOK signal containing a long string of binary 0s: in this case, m = 0 for as long as the data remains 0. The tag will receive no power during this time. If the data remains 0 for too long, the tag will power off and need to be restarted, a situation not likely to be conducive to reliable operation. Even when some binary 1s are present, the power level delivered to the tag is strongly data dependent, an undesirable trait.

A common solution to the power problem is to code the binary data prior to modulation. One RFID coding approach is known as pulse-interval encoding (PIE). A binary '1' is coded as a short power-off pulse following a long full-power interval, and a binary '0' is coded as a shorter full-power interval with the same power-off pulse (Figure 7). The resulting coded baseband signal m(t) is then used to modulate the carrier (Figure 8). PIE using equal low and high pulses for a 0 ensures that at least 50% of the maximum power is delivered to the tag even when the data being transmitted contains long strings of zeros, and if the high is three times as long for a '1', a random stream of equally mixed binary data will provide about 63% of peak power. Note that in this case, the data rate becomes dependent on the data: a stream of binary 0s will be transmitted more rapidly than a stream of binary 1s. A single symbol has two features--the off-time and on-time--but still conveys only one binary bit. (This scheme is used in EPCglobal Class 1 Generation 2 readers. Other passive RFID standards use slightly different coding schemes, all generally characterized by the desire to have the reader power on as much as possible to power the tag.)

Figure 7. Pulse-interval Coding Baseband Symbols (the function m(t)).

Figure 8. Pulse-interval Coding with OOK Modulation of a Carrier Wave

In fixing the problem with transmitted power by replacing OOK with PIE, weve made another problem worse. Radio waves travel everywhere, so in some sense the radio medium is shared between various users. For example, I would like to be able to read tags on packages in my storeroom despite the fact that the storeroom is also illuminated by the local broadcast radio and television stations, cellular phone basestations, the radio link from the taxi across the street, and the satellite downlink to the neighborhood cable TV system. Using a single medium for many signals is known as multiplexing. The most common form of multiplexing in radio, in use for almost a century, is frequency-division multiple access (FDMA): different users transmit using different carrier frequencies, and receivers are adapted to capture only the frequency of interest. (Signals can also be multiplexed in time and in coding. In RFID, time multiplexing is used when a reader uses an anticollision algorithm to poll tags one at a time; see Chapter 8 for more details.) We will discuss the means used to filter the desired frequencies from the received signal in more detail in Chapter 4; for the present, it suffices to know that this operation can be accomplished. An RFID reader transmits on a frequency within the band at 902-928 MHz (in the United States), and listens to responses only within that band, rejecting the AM radio broadcast at 1 MHz, the television transmission at 52 MHz, the cellular transmission at 874 MHz, and so on.

This scheme would seem to allow an unlimited number of users to share the electromagnetic spectrum. However, recall that a signal must be modulated in order to convey information. When we modulate the signal, we increase the signal bandwidth. We saw an indication that this would be so in examining analog sinusoidal modulation of a signal (Figure 5). A modulated signal occupies a finite region of frequency, and neighbors must be separated by something like that amount in frequency to avoid interference.

Furthermore, choices we make in modulation affect how much bandwidth we use. For example, if we modulate the signal faster by making the individual symbols take less time--that is, if we increase the data rate--we use more bandwidth. This phenomenon is illustrated in Figure 9¹, where we show the power spectrum of a modulated signal, and we have made use of the dB notation for spectral power introduced in Section 3.2 above. The spectrum has its largest power near the carrier frequency fc, but a considerable amount of power is transmitted at frequencies rather far from the carrier, as we might have suspected from Figure 3.5 above. The distance from the carrier frequency to the first major 'dip' in the spectrum is inversely proportional to the symbol time t that is, it is the same as the data rate R = 1/t for OOK. The shorter the symbol time, the faster we can send data, but the more bandwidth we use.

How we send symbols also matters. An abrupt step at the edge of each symbol gives more power far from the carrier than a smooth transition between low and high power states, as depicted in Figure 10. (Note that the residual power shown far from the carrier for the smooth symbols in this figure is affected by the specific method of smoothing the symbol and the accuracy of the numerical model.) Of course, the ability to smooth the transitions is limited by the duration of the symbols: at some point, changes happen so slowly that fully on or fully off states are never reached, causing the transmitted power to fall (and become data dependent). Smoothing the signals also makes the receivers problem harder. It doesn't really matter when you test the voltage of a signal like that in left side of Figure 10 as long as you are within the symbol, but the smoothed signal on the right side is best sampled exactly at the center of the symbol, where the power is either at its maximum value or nearly zero. Sampling at any other times will result in more power for a nominal 0 or less power for a nominal '1': that is, the measured modulation depth is reduced. Thus, the receiver needs to do a better job of synchronizing with the incoming signal if that signal is smoothed.

Figure 9. Faster Modulation = Wider Spectrum.

Finally, the way we code the signal also matters. By examination of Figure 6 and Figure 8, we can see that pulse interval encoding will result in shorter pulses than OOK for the same data rate, so from Figure 9, it seems likely that PIE would have a wider spectrum than OOK for the same data rate. This expectation is confirmed in Figure 11: substituting a stream of PIE symbols at the same average data rate for OOK symbols results in reduced power very near the carrier, but more power far from the carrier. In particular, a strong, narrow emission is seen at a frequency which turns out to correspond to (1/duration of a binary '0'); as depicted by the inset in the figure, the strong resemblance of a '0' symbol to a sine function results in a concentration of power at the corresponding frequency. The more diffuse band at half this offset results from the binary '1' symbol.

Figure 10. Abrupt Symbols Have More Power at Frequencies Far From the Carrier. (The Exact Levels Shown Here are Somewhat Dependent on the Modeling Algorithm).

To clarify why this sort of thing matters in real applications, let's look at a practical example. In the United States, unlicensed readers randomly hop from one frequency to another within the ISM band from 902-928 MHz. Typically, RFID readers use channels that are 500 kHz wide and separated by 500 kHz. When a reader is trying to hear a tag, it transmits a signal of constant amplitude and phase. If reader #1on channel 10 is trying to hear a tag, while reader #2 on channel 11 is producing an emission spectra like those shown in Figure 11, the situation would look something like Figure 12, where the spectrum from reader #2 is scaled for a data rate of about 100 kbps and a distance of about 20 m. In Section 3.5 below, we will find that for typical distances, a tag signal is likely to be 40-90 dB smaller than the CW signal from the reader. The leakage from reader #2 into reader #1s channel is thus comparable to or even larger than the tag signal; it will be difficult to detect the tag when reader #2 is transmitting data. Note, this is happening despite the fact that the tags are only 1-3 m from the reader, much closer than the interfering reader!

Even worse, if one of the readers happened to be near the edge of the ISM band, some of this power may be radiated outside of the allowed frequency range, potentially interfering with users of licensed frequencies, who have often paid for the privilege of exclusive use of said spectrum and get upset when they encounter freeloaders. In the United States, the FCC requires that all radios be tested to ensure that such out-of-band radiation is minimized. Interference and out-of-band emissions represent important limits on how fast data can be transmitted by a reader, and on coding and modulation used, because the speed and method of modulation determine the bandwidth of the resulting signal.

Figure 11. Coding Data as PIE Produces a Strong Narrow Emission Far from the Carrier, as Well as a Higher Average Signal Power Far from the Carrier; the inset Shows How This Band Arises from the '0' Symbol. (The Exact Position of These Features Relative to the Data Rate Varies Depending on the Duration of a Binary '1'.)

Let us pause for a bit of mathematics to clarify the frequency scales of the figures above. An ideal abrupt pulse (an OOK binary '1') of duration t has a spectrum:

Figure 12. Power Far from the Carrier of Reader #2 is in the Channel of Reader #1 if Data Rate is High and Unsmoothed PIE is Used.

This function has some useful special values:

In particular, the first zero of this function is at a frequency of (1/t), where t is the duration of the pulse. When the signal is a modulated carrier wave, the spectrum is centered around the carrier frequency, and the zeros are displaced from the carrier by (1/t) (Figure 13).

A stream of binary pulses--an OOK signal as in Figure 6--is just the sum of a number of these pulses, each with the same spectrum, so the full data stream will also have a spectrum with zero value at the same frequency offset from the carrier. These first zeros determine the width of the main lobe of the signal spectrum and are indicated by the dashed lines in Figure 9. Most of the power in the spectrum is contained within the region about half this wide, that is within a frequency range of (fc -1/(2t)) to (fc +1/(2t)). Thus, the narrowest channel that makes sense for an OOK signal is about twice as wide as the inverse of the data rate; we need 200 kHz to fit in 100 kbps.

Figure 13. A Pulse-modulated Carrier and Corresponding Power Spectrum; Square-root of Power is Shown for Clarity..

PIE is much less efficient because the shortest pulse--the high part of a binary '0', Figure 8--is about 1/3 as long as an OOK pulse for the same data rate, so roughly three times as much spectrum is needed. To fit the main lobe of the spectrum within a 500 kHz channel, we can only use a data rate of around 85 kbps--which, as we will see in Chapter 8, is just about the upper limit on reader data rates in United States operation, using unfiltered PIE-like modulations.

To summarize:

• To convey information on a signal, the signal must be modulated.
• Modulation causes the signal's spectrum to expand, requiring allocation of bandwidth in order to avoid interference.
• The peculiar requirements of passive RFID lead to modulation and coding of binary data that are relatively inefficient in spectral use, limiting reader data rates.

It is important to note that more sophisticated radio systems, such as cellular telephony or IEEE 802.11 (WiFi), use modulation techniques that are substantially more efficient users of spectrum than PIE or OOK. However, these methods generally depend on the ability of the receiver to detect changes in the phase of the high-frequency signal rather than simply determining the power level, which passive RFID tags generally cannot do. As we will discuss in more detail in Chapters 4 and 8, single sideband (SSB) and phase-reversal ASK (PR-ASK) modulations, which use phase information at the reader but require only amplitude detection from the tag, can be used to improve the spectral efficiency.

¹It is worth noting that in this and the next few figures, the spectra are calculated for a series of about 80 random data bits, only a few of which are shown in the upper 'signal' display, in order to keep the diagrams intelligible. If we calculated the frequency spectra over a larger number of bits, they would be smoother, but the spectra shown are reasonably representative of the kind of data actually obtained when the output of a typical frequency-hopping RFID reader is examined over short time scales.

Next: Backscatter Radio Links

About the Author
Daniel Dobkin is an RFID consultant, writer and teacher. He holds six patents as inventor or co-inventor. He is the author of such books as: Principles of Chemical Vapor Deposition and RF Engineering for Wireless Networks, and The RF in RFID. Additionally, he is a published author of 25 technical publications. He has taught RFID courses internationally in Singapore for the SMa/RFID Focus; and domestically at SDForum, Mitre Corporate University, and San Jose State University. Daniel is a Stanford University PhD in Applied Physics. He has MS and BS degrees from CalTech.

Printed with permission from Newnes, a Division of Elsevier. Copyright 2007. "The RF in RFID: Passive UHF RFID in Practice by Daniel M. Dobkin. ISBN-10: 0750682094For more information about this title and other similar books, please visit www.elsevierdirect.com