Wireless methods and devices employing steganography Number:7,522,728 from the United States Patent and Trademark Office (PTO) owispatent A wireless device includes a data capture system, a radiant-energy data transmission system, and a steganographic encoder that hides a plural-bit auxiliary code within data captured by the data capture system prior to its transmission by the data transmission system. An illustrative system, operable with audio input data, is a cell phone that steganographically encodes a user's voice.<H steganography, employing, Wireless, method, 7522728.html, Patent, pto, 7522728

Title: Wireless methods and devices employing steganography

Abstract: A wireless device includes a data capture system, a radiant-energy data transmission system, and a steganographic encoder that hides a plural-bit auxiliary code within data captured by the data capture system prior to its transmission by the data transmission system. An illustrative system, operable with audio input data, is a cell phone that steganographically encodes a user's voice.

Patent Number: 7,522,728 Issued on 04/21/2009 to Rhoads

Inventors: Rhoads; Geoffrey B. (West Linn, OR)
Assignee: Digimarc Corporation (Beaverton, OR)

Appl. No.: 09/479,304

Filed: January 6, 2000

Related U.S. Patent Documents


Application Number	Filing Date	Patent Number	Issue Date
09172324	Oct., 1998	6064737
08637531	Apr., 1996	5822436
08534005	Sep., 1995	5832119
08512993	Aug., 1995
08508083	Jul., 1995	5841978
09172324
08438159	May., 1995	5850481
08327426	Oct., 1994	5768426
08215289	Mar., 1994
08154866	Nov., 1993
08438159
PCT/US94/13366	Nov., 1994

Current U.S. Class: 380/270 ; 370/252; 380/247; 380/248; 380/255; 380/28; 455/410; 455/419; 713/150; 713/168; 713/176

Current International Class: H04K 1/00 (20060101)

Field of Search: 380/252,253,28,247-250,255,270 379/100.13 455/403,231,258,270,272,278,410,411,419,420 713/150,168-171,176-181

References Cited [Referenced By]

U.S. Patent Documents


3004104	October 1961	Hembrooke
3406344	October 1968	Hopper
3517315	June 1970	Malm
3586781	June 1971	Jones
3984624	October 1976	Waggener
4225967	September 1980	Miwa
4237484	December 1980	Brown et al.
4238849	December 1980	Gassmann
4313197	January 1982	Maxemchuk
4367488	January 1983	Leventer
4379947	April 1983	Warner
4380027	April 1983	Leventer
4395600	July 1983	Lundy
4425642	January 1984	Moses
4425661	January 1984	Moses et al.
4528588	July 1985	Lofberg
4547804	October 1985	Greenberg
4672605	June 1987	Hustig
4675746	June 1987	Tetrick et al.
4703476	October 1987	Howard
4750173	June 1988	Bluthgen
4807031	February 1989	Broughton
4855827	August 1989	Best
4876617	October 1989	Best et al.
4908836	March 1990	Rushforth
4931871	June 1990	Kramer
4943973	July 1990	Werner
4969041	November 1990	O'Grady
4972471	November 1990	Gross et al.
4979210	December 1990	Nagata et al.
5010405	April 1991	Schreiber et al.
5025443	June 1991	Gupta
5084891	January 1992	Ariyavisitakul et al.
5103459	April 1992	Gilhousen
5113437	May 1992	Best et al.
5134719	July 1992	Mankovitz
5144649	September 1992	Zicker et al.
5146457	September 1992	Veldhuis et al.
5153919	October 1992	Reeds, III et al.
5161210	November 1992	Druyvesteyn et al.
5185762	February 1993	Schilling
5200822	April 1993	Bronfin et al.
5204902	April 1993	Reeds et al.
5243423	September 1993	DeJean
5319735	June 1994	Preuss et al.
5327237	July 1994	Gerdes et al.
5335278	August 1994	Matchett et al.
5345595	September 1994	Johnson et al.
5355161	October 1994	Bird
5379345	January 1995	Greenberg
5387941	February 1995	Montgomery et al.
5388212	February 1995	Grube et al.
5404377	April 1995	Moses
5410541	April 1995	Hotto
5420910	May 1995	Rudokas et al.
5430760	July 1995	Dent
5448760	September 1995	Frederick
5450490	September 1995	Jensen et al.
5454027	September 1995	Kennedy et al.
5465387	November 1995	Mukherjee
5467169	November 1995	Morikawa
5473631	December 1995	Moses
5481611	January 1996	Owens et al.
5513245	April 1996	Mizikovsky et al.
5559559	September 1996	Jungo et al.
5574962	November 1996	Fardeau
5579124	November 1996	Aijala et al.
5584070	December 1996	Harris
5612974	March 1997	Astrachan
5646997	July 1997	Barton
5652626	July 1997	Kawakami et al.
5659726	August 1997	Sandford, II et al.
5663766	September 1997	Sizer
5687191	November 1997	Lee et al.
5712920	January 1998	Spille
5719937	February 1998	Warren
5721788	February 1998	Powell
5761606	June 1998	Wolzien
5764763	June 1998	Jensen et al.
5764770	June 1998	Schipper
5774452	June 1998	Wolosewicz
5822360	October 1998	Lee et al.
5822432	October 1998	Moskowitz et al.
5826227	October 1998	Jayant
5862260	January 1999	Rhoads
5872531	February 1999	Johnson
5905800	May 1999	Moskowitz et al.
5932863	August 1999	Rathus
5937000	August 1999	Lee
5943615	August 1999	Rose et al.
5978773	November 1999	Hudetz
6064737	May 2000	Rhoads
6125172	September 2000	August et al.
6154484	November 2000	Lee
6175627	January 2001	Petrovic et al.
6181802	January 2001	Todd
6278781	August 2001	Rhoads
6282362	August 2001	Murphy et al.
6687345	February 2004	Swartz
2007/0189533	August 2007	Rhoads

Foreign Patent Documents


642060	Apr., 1999	EP
05-236424	Sep., 1993	JP
WO95/20291	Jul., 1995	WO
WO98/27768	Jun., 1998	WO

Other References

Hernandez et al, "Performance Analysis of a 2-D-Multipulse Amplitude Modulation Scheme for Data Hiding and Watermarking of Still Images", IEEE Journal on Selected Areas in Communications, vol. 15, No. 4, May 1998, pp. 510-524. cited by examiner .
Unknown author, "Dynamics (music)--Wikipedia, the free encyclopedia", web page printed on Jun. 26, 2008, 7 pages. cited by examiner .
Szepanski, Additive binary Data Transmission for Video Signals, Conference of the Communications Engineering Society, 1980, NTG Technical Reports, vol. 74, pp. 343-351. (German text and English translation enclosed). cited by other .
Arachelian, "White Noise Storm," Apr. 11, 1994, Internet reference, 13 pages. cited by other .
Bender, "Techniques for Data Hiding," SPIE vol. 2420, pp. 164-173, Jan. 1995. cited by other .
Tirkel et al, "Electronic Water Mark," DICTA-93, Macquarie University, Sydney, Australia, pp. 666-673, Dec. 1993. cited by other .
Van Schyndel et al, "A Digital Watermark," IEEE International Conference on Image Processing, pp. 86-90, Nov. 13-16, 1994. cited by other .
Cox et al, "Secure Spread Spectrum Watermarking for Multimedia," NEC Technical Report, Dec. 1995. cited by other.

Primary Examiner: Vu; Kimyen
Assistant Examiner: Shan; April Y

Claims

I claim:

1. A method of operating a cell phone, comprising: receiving auxiliary data wirelessly sent to the cell phone from a remote transmitter; thereafter, receiving input information and expressing the input information in digital form; steganographically encoding the input information to hide a plural-bit auxiliary code therein, the encoding depending, at least in part, on the received auxiliary data sent to the cell phone from the remote transmitter; and transmitting the steganographically-encoded information from the cell phone by wireless in a digital format; wherein: the input information is digitally marked with the plural-bit auxiliary code prior to being transmitted, but such code is hidden due to its steganographic nature; the plural-bit auxiliary code is retrievable only by entities that (1) receive the information sent by the cell phone; (2) know that a code is present despite being hidden, and (3) have information about its manner of encoding; and the method thereby provides a hidden channel through which a code dependent on auxiliary data earlier sent to a cell phone can be transmitted from the cell phone.

2. The method of claim 1 in which the input information is audio information.

3. The method of claim 1 wherein the steganographic encoding includes additively combining samples of a digital overlay signal with corresponding samples of the input information, the overlay signal being dependent both on the plural-bit auxiliary code and on the input information.

4. The method of claim 1 that further includes wirelessly communicating an identifier from the cell phone, wherein said plural-bit auxiliary code is at least partially redundant with said identifier, so that at least part of said identifier is sent from the cell phone in two different manners.

5. The method of claim 1 wherein said plural-bit auxiliary code comprises an identifier uniquely identifying the cell phone, rather than identifying the input information or a user of cell phone.

6. A cell phone including a data capture system and a radiant-energy transmission system, characterized in that the cell phone further includes a steganographic encoder that modifies data captured by the data capture system in accordance with an encoding signal, to hide a plural-bit auxiliary code within the data prior to transmission by the data transmission system, the steganographic encoder being adapted to generate an encoding signal that depends--in part--on dynamics of the data, wherein data captured by the data capture system is digitally marked with the encoding signal prior to being transmitted by the transmission system, and the steganographic encoder is adapted to generate an encoding signal that is responsive to a first-, second- or higher-order derivative of the data.

7. The cell phone of claim 6 in which the steganographic encoder is adapted to control an amplitude of the encoding signal, in part, in accordance with dynamics of the data.

8. The cell phone of claim 6 further comprising wireless receiver circuitry that provides information to a memory, wherein the steganographic encoder is adapted to generate an encoding signal that depends, in part, on the information in the memory.

9. The cell phone of claim 6 in which the data comprises a series of samples, and the steganographic encoder is adapted to generate an encoding signal that depends on the dynamics of several samples.

Description

RELATED APPLICATION DATA

This application is a division of Ser. No. 09/172,324 (now U.S. Pat. No. 6,064,737) filed on Oct. 13, 1998, which is a division of Ser. No. 08/637,531 (now U.S. Pat. No. 5,822,436) filed on Apr. 25, 1996, which is a continuation-in-part of Ser. No. 08/534,005 (now U.S. Pat. No. 5,832,119) filed on Sep. 25, 1995, which is a continuation-in-part of Ser. No. 08/512,993 (now abandoned) filed on Aug. 9, 1995, which is a continuation-in-part of Ser. No. 08/508,083 (now U.S. Pat. No. 5,841,978) filed on Jul. 27, 1995. Application Ser. No. 09/172,324 is also a continuation-in-part of application Ser. No. 08/438,159, filed May 8, 1995, which is a continuation-in-part of Ser. No. 08/327,426 (now U.S. Pat. No. 5,768,426) filed on Oct. 21, 1994, which is a continuation-in-part of Ser. No. 08/215,289 (now abandoned) filed on Mar. 17, 1994, which is a continuation-in-part of Ser. No. 08/154,866 (now abandoned) filed on Nov. 18, 1993. Application Ser. No. 08/438,159 is also a continuation-in-part of PCT/US94/13366, filed on Nov. 16, 1994. These applications are each incorporated by reference.

TECHNICAL FIELD

The present invention relates to wireless communication systems, such as cellular systems and PCS systems, and more particularly relates to methods involving use of steganographically encoded data in conjunction with such systems.

BACKGROUND AND SUMMARY OF THE INVENTION

(For expository convenience, this disclosure generally refers to cellular telephony systems. However, it should be recognized that the invention is not so limited, but can be used with any wireless communications device, whether for voice or data; analog or digital.)

In the cellular telephone industry, hundreds of millions of dollars of revenue is lost each year through theft of services. While some services are lost due to physical theft of cellular telephones, the more pernicious threat is posed by cellular telephone hackers.

Cellular telephone hackers employ various electronic devices to mimic the identification signals produced by an authorized cellular telephone. (These signals are sometimes called authorization signals, verification numbers, signature data, etc.) Often, the hacker learns of these signals by eavesdropping on authorized cellular telephone subscribers and recording the data exchanged with the cell cite. By artful use of this data, the hacker can impersonate an authorized subscriber and dupe the carrier into completing pirate calls.

In the prior art, identification signals are segregated from the voice signals. Most commonly, they are temporally separated, e.g. transmitted in a burst at the time of call origination. Voice data passes through the channel only after a verification operation has taken place on this identification data. (Identification data is also commonly included in data packets sent during the transmission.) Another approach is to spectrally separate the identification, e.g. in a spectral subband outside that allocated to the voice data.

Other fraud-deterrent schemes have also been employed. One class of techniques monitors characteristics of a cellular telephone's RF signal to identify the originating phone. Another class of techniques uses handshaking protocols, wherein some of the data returned by the cellular telephone is based on an algorithm (e.g. hashing) applied to random data sent thereto.

Combinations of the foregoing approaches are also sometimes employed.

U.S. Pat. Nos. 5,465,387, 5,454,027, 5,420,910, 5,448,760, 5,335,278, 5,345,595, 5,144,649, 5,204,902, 5,153,919 and 5,388,212 detail various cellular telephone systems, and fraud deterrence techniques used therein. The disclosures of these patents are incorporated by reference.

As the sophistication of fraud deterrence systems increases, so does the sophistication of cellular telephone hackers. Ultimately, hackers have the upper hand since they recognize that all prior art systems are vulnerable to the same weakness: the identification is based on some attribute of the cellular telephone transmission outside the voice data. Since this attribute is segregated from the voice data, such systems will always be susceptible to pirates who electronically "patch" their voice into a composite electronic signal having the attribute(s) necessary to defeat the fraud deterrence system.

To overcome this failing, one embodiment steganographically encodes the voice signal with identification data, resulting in "in-band" signaling (in-band both temporally and spectrally). This approach allows the carrier to monitor the user's voice signal and decode the identification data therefrom.

In one form of the invention, some or all of the identification data used in the prior art (e.g. data transmitted at call origination) is repeatedly steganographically encoded in the user's voice signal as well. The carrier can thus periodically or a periodically check the identification data accompanying the voice data with that sent at call origination to ensure they match. If they do not, the call is identified as being hacked and steps for remediation can be instigated such as interrupting the call.

In another form of the invention, a randomly selected one of several possible messages is repeatedly steganographically encoded on the subscriber's voice. An index sent to the cellular carrier at call set-up identifies which message to expect. If the message steganographically decoded by the cellular carrier from the subscriber's voice does not match that expected, the call is identified as fraudulent.

In certain embodiments, the steganographic encoding relies on a pseudo random data signal to transform the message or identification data into a low level noise-like signal superimposed on the subscriber's digitized voice signal. This pseudo random data signal is known, or knowable, to both the subscriber's telephone (for encoding) and to the cellular carrier (for decoding). Many such embodiments rely on a deterministic pseudo random number generator seeded with a datum known to both the telephone and the carrier. In simple embodiments this seed can remain constant from one call to the next (e.g. a telephone ID number). In more complex embodiments, a pseudo-one-time pad system may be used, wherein a new seed is used for each session (i.e. telephone call). In a hybrid system, the telephone and cellular carrier each have a reference noise key (e.g. 10,000 bits) from which the telephone selects a field of bits, such as 50 bits beginning at a randomly selected offset, and each uses this excerpt as the seed to generate the pseudo random data for encoding. Data sent from the telephone to the carrier (e.g. the offset) during call set-up allows the carrier to reconstruct the same pseudo random data for use in decoding. Yet further improvements can be derived by borrowing basic techniques from the art of cryptographic communications and applying them to the steganographically encoded signal detailed in this disclosure.

Details of applicant's preferred techniques for steganographic encoding/decoding with a pseudo random data stream are more particularly detailed in applicant's prior applications, but the present invention is not limited to use with such techniques. A brief review of other steganographic techniques suitable for use with the present invention follows.

British patent publication 2,196,167 to Thorn EMI discloses a system in which an audio recording is electronically mixed with a marking signal indicative of the owner of the recording, where the combination is perceptually identical to the original. U.S. Pat. Nos. 4,963,998 and 5,079,648 disclose variants of this system.

U.S. Pat. No. 5,319,735 to B.B.N. rests on the same principles as the earlier Thorn EMI publication, but additionally addresses psycho-acoustic masking, issues.

U.S. Pat. Nos. 4,425,642, 4,425,661, 5,404,377 and 5,473,631 to Moses disclose various systems for imperceptibly embedding data into audio signals--the latter two patents particularly focusing on neural network implementations and perceptual coding details.

U.S. Pat. No. 4,943,973 to AT&T discloses a system employing spread spectrum techniques for adding a low level noise signal to other data to convey auxiliary data therewith. The patent is particularly illustrated in the context of transmitting network control signals along with digitized voice signals.

U.S. Pat. No. 5,161,210 to U.S. Philips discloses a system in which additional low-level quantization levels are defined on an audio signal to convey, e.g., a copy inhibit code, therewith.

U.S. Pat. No. 4,972,471 to Gross discloses a system intended to assist in the automated monitoring of audio (e.g. radio) signals for copyrighted materials by reference to identification signals subliminally embedded therein.

There are a variety of shareware programs available on the internet (e.g. "Stego" and "White Noise Storm") which generally operate by swapping bits from a to-be-concealed message stream into the least significant bits of an image or audio signal. White Noise Storm effects a randomization of the data to enhance its concealment.

A British company, Highwater FBI, Ltd., has introduced a software product which is said to imperceptibly embed identifying information into photographs and other graphical images. This technology is the subject of European patent applications 9400971.9 (filed Jan. 19, 1994), 9504221.2 (filed Mar. 2, 1995), and 9513790.7 (filed Jul. 3, 1995), the first of which has been laid open as PCT publication WO 95/20291.

Walter Bender at M.I.T. has done a variety of work in the field, as illustrated by his paper "Techniques for Data Hiding," Massachusetts Institute of Technology, Media Laboratory, January 1995.

Dice, Inc. of Palo Alto has developed an audio marking technology marketed under the name Argent. While a U.S. patent application is understood to be pending, it has not yet been issued.

Tirkel et al, at Monash University, have published a variety of papers on "electronic watermarking" including, e.g., "Electronic Water Mark," DICTA-93, Macquarie University, Sydney, Australia, December, 1993, pp. 666-673, and "A Digital Watermark," IEEE International Conference on Image Processing, Nov. 13-16, 1994, pp. 86-90.

Cox et al, of the NEC Technical Research Institute, discuss various data embedding techniques in their published NEC technical report entitled "Secure Spread Spectrum Watermarking for Multimedia," December, 1995.

Moller et al. discuss an experimental system for imperceptibly embedding auxiliary data on an ISDN circuit in "Rechnergestutzte Steganographie: Wie sie Funktioniert und warum folglich jede Reglementierung von Verschlusselung unsinnig ist," DuD, Datenschutz und Datensicherung, 18/6 (1994) 318-326. The system randomly picks ISDN signal samples to modify, and suspends the auxiliary data transmission for signal samples which fall below a threshold.

In addition to the foregoing, many of the other cited prior art patents and publications disclose systems for embedding a data signal on an audio signal. These, too, can generally be employed in systems according to the present invention.

The foregoing and additional features and advantages of certain embodiments of the present invention will be more readily apparent from the following detailed description, which proceeds with reference to the accompanying drawings.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a block diagram showing principal components of an exemplary wireless telephony system.

FIG. 2 is a block diagram of an exemplary steganographic encoder that can be used in the telephone of the FIG. 1 system.

FIG. 3 is a block diagram of an exemplary steganographic decoder that can be used in the cell site of the FIG. 1 system.

FIGS. 4A and 4B are histograms illustrating signal relationships which may be exploited to facilitate decoding.

DETAILED DESCRIPTION

The reader is presumed to be familiar with cellular communications technologies. Accordingly, details known from prior art in this field aren't belabored herein.

Referring to FIG. 1, an illustrative cellular system includes a telephone 10, a cell site 12, and a central office 14.

Conceptually, the telephone may be viewed as including a microphone 16, an A/D converter 18, a data formatter 20, a modulator 22, an RF section 24, an antenna 26, a demodulator 28, a data unformatter 30, a D/A converter 32, and a speaker 34.

In operation, a subscriber's voice is picked up by the microphone 16 and converted to digital form by the A/D converter 18. The data formatter 20 puts the digitized voice into packet form, adding synchronization and control bits thereto. The modulator 22 converts this digital data stream into an analog signal whose phase and/or amplitude properties change in accordance with the data being modulated. The RF section 24 commonly translates this time-varying signal to one or more intermediate frequencies, and finally to a UHF transmission frequency. The RF section thereafter amplifies it and provides the resulting signal to the antenna 26 for broadcast to the cell site 12.

The process works in reverse when receiving. A broadcast from the cell site is received through the antenna 26. RF section 24 amplifies and translates the received signal to a different frequency for demodulation. Demodulator 28 processes the amplitude and/or phase variations of the signal provided by the RF section to produce a digital data stream corresponding thereto. The data unformatter 30 segregates the voice data from the associated synchronization/control data, and passes the voice data to the D/A converter for conversion into analog form. The output from the D/A converter drives the speaker 34, through which the subscriber hears the other party's voice.

The cell site 12 receives broadcasts from a plurality of telephones 10, and relays the data received to the central office 14. Likewise, the cell site 12 receives outgoing data from the central office and broadcasts same to the telephones.

The central office 14 performs a variety of operations, including call authentication, switching, and cell hand-off.

(In some systems, the functional division between the cell site and the central station is different than that outlined above. Indeed, in some systems, all of this functionality is provided at a single site.)

In an exemplary embodiment of the present invention, each telephone 10 additionally includes a steganographic encoder 36. Likewise, each cell site 12 includes a steganographic decoder 38. The encoder operates to hide an auxiliary data signal among the signals representing the subscriber's voice. The decoder performs the reciprocal function, discerning the auxiliary data signal from the encoded voice signal. The auxiliary signal serves to verify the legitimacy of the call.

An exemplary steganographic encoder 36 is shown in FIG. 2.

The illustrated encoder 36 operates on digitized voice data, auxiliary data, and pseudo-random noise (PRN) data. The digitized voice data is applied at a port 40 and is provided, e.g., from A/D converter 18. The digitized voice may comprise 8-bit samples. The auxiliary data is applied at a port 42 and comprises, in one form of the invention, a stream of binary data uniquely identifying the telephone 10. (The auxiliary data may additionally include administrative data of the sort conventionally exchanged with a cell site at call set-up.) The pseudo-random noise data is applied at a port 44 and can be, e.g., a signal that randomly alternates between "-1" and "1" values. (More and more cellular phones are incorporating spread spectrum capable circuitry, and this pseudo-random noise signal and other aspects of this invention can often "piggy back" or share the circuitry which is already being applied in the basic operation of a cellular unit).

For expository convenience, it is assumed that all three data signals applied to the encoder 36 are clocked at a common rate, although this is not necessary in practice.

In operation, the auxiliary data and PRN data streams are applied to the two inputs of a logic circuit 46. The output of circuit 46 switches between -1 and +1 in accordance with the following table:

TABLE-US-00001 AUX PRN OUTPUT 0 -1 1 0 1 -1 1 -1 -1 1 1 1

(If the auxiliary data signal is conceptualized as switching between -1 and 1, instead of 0 and 1, it will be seen that circuit 46 operates as a one-bit multiplier.)

The output from gate 46 is thus a bipolar data stream whose instantaneous value changes randomly in accordance with the corresponding values of the auxiliary data and the PRN data. It may be regarded as noise. However, it has the auxiliary data encoded therein. The auxiliary data can be extracted if the corresponding PRN data is known.

The noise-like signal from gate 46 is applied to the input of a scaler circuit 48. Scaler circuit scales (e.g. multiplies) this input signal by a factor set by a gain control circuit 50. In the illustrated embodiment, this factor can range between 0 and 15. The output from scaler circuit 48 can thus be represented as a five-bit data word (four bits, plus a sign bit) which changes each clock cycle, in accordance with the auxiliary and PRN data, and the scale factor. The output from the scaler circuit may be regarded as "scaled noise data" (but again it is "noise" from which the auxiliary data can be recovered, given the PRN data).

The scaled noise data is summed with the digitized voice data by a summer 51 to provide the encoded output signal (e.g. binarily added on a sample by sample basis). This output signal is a composite signal representing both the digitized voice data and the auxiliary data.

The gain control circuit 50 controls the magnitude of the added scaled noise data so its addition to the digitized voice data does not noticeably degrade the voice data when converted to analog form and heard by a subscriber. The gain control circuit can operate in a variety of ways.

One is a logarithmic scaling function. Thus, for example, voice data samples having decimal values of 0, 1 or 2 may be correspond to scale factors of unity, or even zero, whereas voice data samples having values in excess of 200 may correspond to scale factors of 15. Generally speaking, the scale factors and the voice data values correspond by a square root relation. That is, a four-fold increase in a value of the voice data corresponds to approximately a two-fold increase in a value of the scaling factor associated therewith. Another scaling function would be linear as derived from the average power of the voice signal.

(The parenthetical reference to zero as a scaling factor alludes to cases, e.g., in which the digitized voice signal sample is essentially devoid of information content.)

More satisfactory than basing the instantaneous scaling factor on a single voice data sample, is to base the scaling factor on the dynamics of several samples. That is, a stream of digitized voice data which is changing rapidly can camouflage relatively more auxiliary data than a stream of digitized voice data which is changing slowly. Accordingly, the gain control circuit 50 can be made responsive to the first, or preferably the second- or higher-order derivative of the voice data in setting the scaling factor.

In still other embodiments, the gain control block 50 and scaler 48 can be omitted entirely.

(Those skilled in the art will recognize the potential for "rail errors" in the foregoing systems. For example, if the digitized voice data consists of 8-bit samples, and the samples span the entire range from 0 to 255 (decimal), then the addition or subtraction of scaled noise to/from the input signal may produce output signals that cannot be represented by 8 bits (e.g. -2, or 257). A number of well-understood techniques exist to rectify this situation, some of them proactive and some of them reactive. Among these known techniques are: specifying that the digitized voice data shall not have samples in the range of 0-4 or 241-255, thereby safely permitting combination with the scaled noise signal; and including provision for detecting and adaptively modifying digitized voice samples that would otherwise cause rail errors.)

Returning to the telephone 10, an encoder 36 like that detailed above is desirably interposed between the A/D converter 18 and the data formatter 20, thereby serving to steganographically encode all voice transmissions with the auxiliary data. Moreover, the circuitry or software controlling operation of the telephone is arranged so that the auxiliary data is encoded repeatedly. That is, when all bits of the auxiliary data have been encoded, a pointer loops back and causes the auxiliary data to be applied to the encoder 36 anew. (The auxiliary data may be stored at a known address in RAM memory for ease of reference.)

It will be recognized that the auxiliary data in the illustrated embodiment is transmitted at a rate one-eighth that of the voice data. That is, for every 8-bit sample of voice data, scaled noise data corresponding to a single bit of the auxiliary data is sent. Thus, if voice samples are sent at a rate of 4800 samples/second, auxiliary data can be sent at a rate of 4800 bits/second. If the auxiliary data is comprised of 8-bit symbols, auxiliary data can be conveyed at a rate of 600 symbols/second. If the auxiliary data consists of a string of even 60 symbols, each second of voice conveys the auxiliary data ten times. (Significantly higher auxiliary data rates can be achieved by resorting to more efficient coding techniques, such as limited-symbol codes (e.g. 5- or 6-bit codes), Huffman coding, etc.) This highly redundant transmission of the auxiliary data permits lower amplitude scaled noise data to be used while still providing sufficient signal-to-noise headroom to assure reliable decoding--even in the relatively noisy environment associated with radio transmissions.

Turning now to FIG. 3, each cell site 12 has a steganographic decoder 38 by which it can analyze the composite data signal broadcast by the telephone 10 to discern and separate the auxiliary data and digitized voice data therefrom. (The decoder desirably works on unformatted data (i.e. data with the packet overhead, control and administrative bits removed; this is not shown for clarity of illustration).

The decoding of an unknown embedded signal (i.e. the encoded auxiliary signal) from an unknown voice signal is best done by some form of statistical analysis of the composite data signal.

In one approach, decoding relies on recombining the composite data signal with PRN data (identical to that used during encoding), and analyzing the entropy of the resulting signal. "Entropy" need not be understood in its most strict mathematical definition, it being merely the most concise word to describe randomness (noise, smoothness, snowiness, etc.).

Most serial data signals are not random. That is, one sample usually correlates--to some degree--with adjacent samples. This is true in sampled voice signals.

Noise, in contrast, typically is random. If a random signal (e.g. noise) is added to (or subtracted from) a non-random signal (e.g. voice), the entropy of the resulting signal generally increases. That is, the resulting signal has more random variations than the original signal. This is the case with the composite data signal produced by encoder 36; it has more entropy than the original, digitized voice data.

If, in contrast, the addition of a random signal to (or subtraction from) a non-random (e.g. voice) signal reduces entropy, then something unusual is happening. It is this anomaly that can be used to decode the composite data signal.

To fully understand this entropy-based decoding method, it is first helpful to highlight a characteristic of the original encoding process: the similar treatment of every Nth (e.g. 480th) sample.

In the encoding process discussed above, the auxiliary data is 480 bits long. Since it is encoded repeatedly, every 480th sample of the composite data signal corresponds to the same bit of the auxiliary data. If this bit is a "1", the scaled PRN data corresponding thereto are added to the digitized voice signal; if this bit is a "0", the scaled PRN data corresponding thereto are subtracted. Due to the repeated encoding of the auxiliary data, every 480th sample of the composite data signal thus shares a characteristic: they are all either augmented by the corresponding noise data (which may be negative), or they are all diminished, depending on whether the bit of the auxiliary data is a "1" or a "0".

To exploit this characteristic, the entropy-based decoding process treats every 480th sample of the composite signal in like fashion. In particular, the process begins by adding to the 1st, 481st, 861st, etc. samples of the composite data signal the PRN data with which these samples were encoded. (That is, a set of sparse PRN data is added: the original PRN set, with all but every 480th datum zeroed out.) The localized entropy of the resulting signal around these points (i.e. the composite data signal with every 480th sample modified) is then computed.

(Computation of a signal's entropy or randomness is well understood by artisans in this field. One generally accepted technique is to take the derivative of the signal at each sample point near a point in question (e.g. the modified sample and 4 samples either side), square these values, and then sum the resulting signals over all of the localized regions over the entire signal. A variety of other well known techniques can alternatively be used.)

The foregoing step is then repeated, this time subtracting the PRN data corresponding thereto from the 1st, 481st, 961st, etc. composite data samples.

One of these two operations will counteract (e.g. undo) the encoding process and reduce the resulting signal's entropy; the other will aggravate it. If adding the sparse PRN data to the composite data reduces its entropy, then this data must earlier have been subtracted from the original voice signal. This indicates that the corresponding bit of the auxiliary data signal was a "0" when these samples were encoded. (A "0" at the auxiliary data input of logic circuit 46 caused it to produce an inverted version of the corresponding PRN datum as its output datum, resulting in subtraction of the corresponding PRN datum from the voice signal.)

Conversely, if subtracting the sparse PRN data from the composite data reduces its entropy, then the encoding process must have earlier added this noise. This indicates that the value of the auxiliary data bit was a "1" when samples 1, 481, 961, etc., were encoded.

By noting in which case entropy is lower by (a) adding or (b) subtracting a sparse set of PRN data to/from the composite data, it can be determined whether the first bit of the auxiliary data is (a) a "0", or (b) a "1." (In real life applications, in the presence of various distorting phenomena, the composite signal may be sufficiently corrupted so that neither adding nor subtracting the sparse PRN data actually reduces entropy. Instead, both operations will increase entropy. In this case, the "correct" operation can be discerned by observing which operation increases the entropy less.)

The foregoing operations can then be conducted for the group of spaced samples of the composite data beginning with the second sample (i.e. 2, 482, 962, . . . ). The entropy of the resulting signals indicate whether the second bit of the auxiliary data signal is a "0" or a "1." Likewise with the following 478 groups of spaced samples in the composite signal, until all 480 bits of the code word have been discerned.

It will be appreciated that the foregoing approach is not sensitive to corruption mechanisms that alter the values of individual samples; instead, the process considers the entropy of spaced excerpts of the composite data, yielding a high degree of confidence in the results.

A second and probably more common decoding technique is based on correlation between the composite data signal and the PRN data. Such operations are facilitated in the present context since the auxiliary data whose encoded representation is sought, is known, at least in large part, a pri