Bobier's 'TriState Integer Cycle Modulation'

One of xG's recent patents is titled "TRI-STATE INTEGER CYCLE MODULATION". The inventor is Joseph Bobier, and the patent number is 7,003,047.

I have made a first pass through this patent and I have to say that it completely belies Mr. Bobier's claims to having made major breakthroughs. Like many would-be inventors of revolutionary modulation methods, he seems to have just enough knowledge to be dangerous. Not only is the method he describes not novel, but it does not and cannot yield the benefits he claims.

As described in column 5, lines 27-43, I quickly recognized his method as just continuous phase frequency shift keying (CPFSK) with biphase (Manchester II) encoding of the baseband data signal. That is, he encodes a '1' by shifting the carrier frequency to a higher frequency, then a lower one, while a '0' is a shift down and then up, or vice versa. These methods have been around for decades, though perhaps not in the exact configuration shown here for the simple reason that it provides no extra benefits.

Quoting from lines 32-36:

These three (3) frequencies can be extremely close (e.g., less than 30 Khz apart, which is a standard cellular channel) or significantly further apart, depending upon the ability of the receiver to differentiate the frequencies. [emphasis added]
In binary frequency shift keying (BFSK) the transmitter sends each bit as one of two radio frequency "tones". FSK can encode two bits as one of four possible tones ("quaternary modulation"), or three bits as one of eight possible tones ("octal"), and so on. Each tone in a sequence of tones is called a "symbol", but in binary modulation a symbol and a bit are the same. The receiver decides which symbol was sent by comparing the received signal with local copies of each tone in a device called a "correlator bank", or by running it through a bank of filters, one tuned to each tone. The correlator or filter with the largest response indicates the tone that was most likely sent.

Although Bobier mentions 3 frequencies, his scheme is really binary -- only two frequencies are used -- if we quite reasonably assume that consecutive data bits are sent back-to-back as there is no reason to return to the carrier frequency between bits.

With only a few, well defined exceptions every efficient digital modulation method must choose its signal set so that the receiver cannot easily mistake one signal for another. One very important type of modulation, binary phase shift keying (BPSK) has an "antipodal" signal set: the signal for a '1' is the inverse of the signal for a '0'. When it comes to telling two signals apart in noise, you can't do any better than this.

Properly implemented, frequency shift keying (FSK) is an orthogonal modulation method. That is, each receiver filter only responds to its own tone and ignores the others. Noise can still cause errors, but without noise this scheme will always give the right answer. Because binary FSK is only orthogonal, not antipodal, it does not perform as well in noise as BPSK.

Bobier says in lines 44-50 that he switches between different frequencies only at the zero crossings, and he sends complete cycles at each frequency. He apparently doesn't realize that merely selecting different tone frequencies, or even sending a complete cycle of each tone, is not sufficient to make them orthogonal. Indeed, he is probably unaware even that orthogonality is necessary. Without it, both filters or correlators will respond to both tones, with the correct one responding only slightly more strongly. In effect, the signal now interferes with itself. It will be much more susceptible to noise, and it might not work even in the absence of noise.

For tones of different frequencies to be orthogonal, an additional requirement must be met: the correlation interval must span an integral number of half cycles of both tones. For example, the interval could span one cycle of tone A and a half cycle of tone B, or 9 cycles of tone A and 10 cycles of tone B. This imposes a firm requirement that the closer the two tones are in frequency, the slower the data rate must be to maintain orthogonality. For example, if the tones are 30 kHz apart as Bobier suggests, then it will take 1/60,000 of a second for the two tones -- whatever their actual frequencies -- to change from being in phase to being 180 degrees (one half cycle) out of phase. This corresponds to a maximum data rate of 60 kb/s. He cannot use such a narrow shift for high speed data.

These same principles control the frequency spacing and data rates of the individual carriers in orthogonal frequency division multiplex, or OFDM, now widely used in systems ranging from 802.11g WiFi to DSL to DVB-T, the digital TV broadcasting method used outside the US.

That FSK requires a frequency shift proportional to the data rate has been known for a very long time. There's even a special name for continuous-phase FSK with the smallest possible shift: Minimum Shift Keying or MSK (not to be confused with the completely unworkable VMSK). MSK is very widely used for sending digital data on band-limited channels such as 2-way police and fire radios.

The other component of Bobier's scheme is the encoding of each data bit as a two-symbol FSK sequence: low, then high; or high, then low. But this is just biphase encoding, also known as Manchester II coding. It's a form of binary phase shift keying that shifts the data spectrum away from zero frequency (DC) so that it may be handled by transformers and capacitor-coupled amplifiers that cannot pass DC or very low frequencies. It comes at the cost of doubling the bandwidth of a given data stream versus simple NRZ coding. Biphase coding used to be very popular; one widespread use is 10Base-T, 10 Mb/s Ethernet. It is still used on many deep space telemetry links to move the data sidebands away from the carrier to make it easier to track. Biphase has since been largely replaced by other spectrum-shaping codes that achieve the desired reduction in low frequency signal components with less additional bandwidth. One example is Eight-to-Fourteen Modulation or EFM, used on the Compact Disc to keep low frequency audio signals from interfering with the laser tracking servos.


Bobier makes other comments in this patent that demonstrate a surprising ignorance of the basic principles of radio communication. For example in column 6 lines 29-32 he says:
Inasmuch as these zero positions correspond with the absence of electro-magnetic wave energy, no wave disturbances are invoked which, would in turn, produce side frequencies. As a consequence, the assigned carrier frequencies may be quite close together in value to provide a substantially improved utilization of the radio spectrum for binary data transmittal.
This is simply wrong! Any change to the amplitude, phase or frequency of a sinusoid generates sidebands no matter when that change is made, even at the zero crossings. By definition, sidebands consist of all of the Fourier components of a modulated signal except for the carrier, which is itself another Fourier component. The instantaneous phase of each Fourier component increases steadily with time according to its frequency so its instantaneous value varies sinusoidally with time, but the amplitude and frequency of each component remains fixed during the entire analysis interval. That's what makes Fourier analysis so elegant and useful: you can represent any time-varying signal as the sum of a whole bunch of sines and cosines, each with a constant amplitude and frequency. Of course it may take many sines and cosines to do this -- perhaps even an infinite number -- especially if your signal changes instantaneously.


The bottom line is very simple: the techniques described in this patent are not novel, having been around for decades. They cannot provide the advantages claimed by the inventor. He seems unaware that it will work very poorly (or more likely, not at all) without proper selection of the frequency shift. If the frequency shift is set to the minimum workable value -- exactly one half the data rate -- then it becomes MSK with Manchester encoding, with identical performance. And the inventor's naive remarks reveal a surprisingly poor understanding of the basic mathematical principles of modern digital communications.

Phil Karn, 16 May 2007