Source-linked AI summary
Hiding Information in Noise: Fundamental Limits of Covert Wireless Communication
Boulat A. Bash, Dennis Goeckel, Saikat Guha, Don Towsley
TL;DR
Conventional security can protect message content without hiding that communication exists, motivating covert wireless links that evade detection. The article reviews steganography and spread-spectrum communication, develops fundamental-limit results, and outlines shadow networks using relays and jammers.
Problem
Conventional cryptography and information-theoretic secrecy protect message content but do not prevent discovery of the communication itself.
Method
The article reviews steganography and spread-spectrum systems, then analyzes fundamental limits for covert communication over AWGN and discrete memoryless channels.
Results
Covert communication generally follows a square-root law, while positive-rate operation can occur when Willie permits informative transmissions or lacks knowledge of his noise statistics.
Takeaways & Limitations
The results motivate future shadow networks combining covert links with relays and friendly jammers that impair wardens’ detection ability.
Abstract
from arXiv · showhide
Widely-deployed encryption-based security prevents unauthorized decoding, but does not ensure undetectability of communication. However, covert, or low probability of detection/intercept (LPD/LPI) communication is crucial in many scenarios ranging from covert military operations and the organization of social unrest, to privacy protection for users of wireless networks. In addition, encrypted data or even just the transmission of a signal can arouse suspicion, and even the most theoretically robust encryption can often be defeated by a determined adversary using non-computational methods such as side-channel analysis. Various covert communication techniques were developed to address these concerns, including steganography for finite-alphabet noiseless applications and spread-spectrum systems for wireless communications. After reviewing these covert communication systems, this article discusses new results on the fundamental limits of their capabilities, as well as provides a vision for the future of such systems.
I. INTRODUCTION
Conventional cryptography protects message content but not the existence of communication, exposing users to detection, metadata collection, and suspicion. The article therefore studies covert point-to-point links, asking how Alice can reliably communicate with Bob while hiding transmission from Willie.
- Encryption and information-theoretic secrecy protect message content but do not prevent adversaries from discovering that communication exists.
- Transmission attempts expose relationships, encrypted traffic can arouse suspicion, and non-computational attacks such as side-channel analysis can defeat cryptographic schemes.
- Covert or LPD/LPI systems aim to prevent detection of transmission attempts while protecting message information.
- The article focuses on the fundamental limits of links that let Alice reliably transmit to Bob while hiding communication from Willie.
- Steganography hides messages in innocuous objects, while physical-layer methods hide transmissions in channel artifacts such as noise.
II. STEGANOGRAPHY
Steganography hides messages in finite-length covertext, but its covert payload is generally constrained by a square-root law unless Alice has a sufficiently accurate covertext model. Its application-layer assumptions limit direct use for physical-layer wireless communication.
- Modern digital steganography embeds messages in finite-length, finite-alphabet covertext, producing stegotext whose altered properties steganalysis seeks to detect.
- O(√n log n) bits can be safely embedded by modifying O(√n) of n covertext symbols, with O(√n log n) secret bits pre-shared with Bob.
- The square-root law yields zero-rate steganography as n grows.
- An empirical covertext model can break the square-root law and achieve positive-rate steganography by producing statistically matching stegotext.
- Steganography has limited physical-layer relevance because it assumes noiseless finite-alphabet channels, replaces covertext, and requires transmitting stegotext when communication may be prohibited.
III. PHYSICAL LAYER COVERT COMMUNICATION
Physical-layer covert communication studies wireless methods that conceal transmissions within radio-channel behavior rather than replacing application-layer covertext. Spread-spectrum techniques are presented as an early physical-layer security approach.
- Spread-spectrum techniques were developed to protect wireless RF communication from detection, jamming, and eavesdropping.
A. Spread Spectrum Communication
Spread-spectrum systems reduce transmitted power spectral density by distributing a signal across bandwidth wider than its original bandwidth. DSSS and FHSS use shared random patterns that enable Bob to recover the signal while remaining secret from Willie.
- A. Spread Spectrum Communication: Spread spectrum transmits a signal requiring bandwidth W_M over much wider bandwidth W_S ≫ W_M, suppressing its power spectral density below the noise floor.
- A. Spread Spectrum Communication: Spread-spectrum systems provide covert communication and resistance to jamming, fading, and interference; typical forms include DSSS and FHSS.
- A. Spread Spectrum Communication: In DSSS, Alice multiplies the signal by a high-bandwidth random binary spreading sequence, and Bob uses the same sequence to de-spread it.
- A. Spread Spectrum Communication: DSSS requires Alice and Bob to exchange the spreading sequence before transmission and keep it secret from Willie.
- A. Spread Spectrum Communication: FHSS retunes the carrier frequency for each symbol using a randomly generated, secretly shared hopping pattern and can combine with OFDM or time hopping.
- A. Spread Spectrum Communication: The article’s fundamental results apply to classical spread-spectrum systems and newer proposals that hide communication in channel noise and equipment imperfections.
B. Square Root Law for Covert Communication over AWGN Channels
Over AWGN channels, covert communication is constrained by Willie’s ability to distinguish Alice’s signal power from noise. The resulting square root law limits reliable covert throughput while secret-sharing schemes support communication under this constraint.
- AWGN channel model: Spread-spectrum signaling reduces Willie’s SNR by distributing signal power across time and frequency, but covert power and reliable information limits remain fundamental questions.The section studies how small Alice’s power must be and how much covert information can be reliably transmitted.
- AWGN channel model: An AWGN model represents free-space RF communication by adding independent zero-mean Gaussian noise to Alice’s transmissions.The model assigns Gaussian noise variances to the links from Alice to Bob and Willie.
- AWGN channel model: Willie’s observations impose a total transmit-power constraint because excess signal power makes Alice’s transmission statistically distinguishable from noise.A standard radiometer suffices for detection when Alice emits more power under the stated noise conditions.
- Square root law: At most O(√n) covert bits can be reliably transmitted in n channel uses, yielding a zero-rate channel as n grows.This square root law parallels digital steganography, while the steganographic setting adds a log n factor because its channel to Bob is noiseless.
- Secret-assisted communication: A construction using arbitrary error-correction codes reliably transmits O(√n) covert bits with O(√n log n) pre-shared secret bits.The secret selects a random subset of channel uses and includes a one-time pad that prevents Willie from exploiting code structure.
- Secret-assisted communication: The pre-shared key is asymptotically larger than the transmitted message, although the paper notes that this trade-off can be acceptable in scenarios where detection is costly.The cited construction’s formal proof is outside the article’s scope.
C. Digital Covert Communication
Digital covert communication extends the square root law from binary symmetric channels to broader discrete memoryless channels using channel resolvability. Secret-less operation is achievable when Willie’s channel is sufficiently worse than Bob’s, while otherwise a sublinear secret can suffice.
- Discrete memoryless channels: A discrete memoryless channel models communication with discrete input and output alphabets whose outputs depend statistically only on the input at the same time.The model includes a designated “no transmission” input available to Willie.
- Binary symmetric channel: On binary symmetric channels, at most O(√n) covert bits are reliably transmitted in n uses, and no pre-shared secret is needed when Willie’s crossover probability exceeds Bob’s.The condition is p_w > p_b.
- Channel resolvability: Channel resolvability generalizes the square root law to discrete memoryless channels by minimizing input entropy needed to approximate a target channel-output distribution.Variational distance and relative entropy are examples of distribution-closeness measures used in this framework.
- Secret-less covert communication: When Willie’s channel is worse than Bob’s, covert communication can follow the square root law without a pre-shared secret.The cited results apply to DMCs and extend to AWGN channels under corresponding noise conditions.
D. Willie’s Ignorance of Transmission Time Helps Alice
When Willie does not know when transmission may occur, Alice can transmit additional information to Bob. Under mild conditions, Alice and Bob may not need to pre-arrange the communication time.
- Unknown transmission timing allows Alice to transmit additional information to Bob.This differs from square-root-law derivations that assume Willie knows the transmission time.
- Under mild conditions relating total available time to transmission duration, Alice and Bob need not pre-arrange the communication time.
E. Positive-rate Covert Communication
Covert communication is generally zero-rate, but positive-rate transmission of O(n) covert bits in n channel uses is possible under specific conditions involving Willie’s knowledge or permissions.
- The article notes that the mathematical analysis of one O(√n)-secret-bit scheme is highly technical and outside its scope.
- O(n) covert bits can be reliably transmitted in n channel uses in positive-rate covert communication.
- Positive rates arise when Willie permits informative transmissions or lacks knowledge of his channel’s probabilistic noise structure.
- When Willie permits transmissions, covert capacity equals information-theoretic secrecy capacity.
- Random noise-power fluctuations can yield positive-rate covert communication in AWGN channels, even when noise power is bounded.
F. Covert Broadcast
Single-warden point-to-point covert communication results can extend to multiple independently controlled receivers when each receiver obtains the required pre-shared secret.
- Point-to-point covert communication results with one warden can extend to multiple independently controlled receivers.
- In AWGN channels, covert communication imposes a power constraint on Alice, while pre-shared secrets enable covert encoding for multiple recipients.
- The multi-warden extension and other networked covert scenarios remain ongoing work.
IV. CONCLUSION: TOWARDS SHADOW NETWORKS
The paper envisions shadow networks built from relays, transmitters, receivers, and friendly jammers that impair wardens’ detection ability. It identifies multi-warden, multi-jammer covert links as an important next step.
- Shadow networks combine relays that handle data with jammers that generate artificial noise to impair wardens’ detection.
- Jammers can be cheap, numerous, and disposable, whereas relays are valuable and require protection.
- Independent jammer activity prevents wardens from detecting relay transmissions by listening to the jammers.
- Extending the model to variable jamming power, multipath fading, multiple wardens, and multiple jammers is identified as important future work.
- Noise uncertainty and variable jamming power may enable positive-rate covert communication.