STEEP: Echoing Encrypted Probes for Secure Messaging

• The paper’s main contribution is adapting probe asymmetry to form an effective return-link wiretap channel that yields a positive secrecy rate even when eavesdroppers have stronger channels.
• STEEP employs a two-phase design with an initial probing phase followed by an echoing phase, enabling Alice to remove probe uncertainty while Eve relies on noisy estimates.
• The method generalizes to MIMO and PSK settings, offering a practical alternative to QKD by using conventional coding and non-regenerative relay technologies.

Searching arXiv for STEEP and closely related secret-message transmission papers. Secret-message Transmission by Echoing Encrypted Probes (STEEP) is a round-trip communication scheme for information-theoretically secure transmission over classical channels. In its canonical form, Alice first sends random probes to Bob, and Bob then echoes back an encrypted function of his probe observation together with a secret message. The central design objective is to create an effective return-link wiretap channel in which Alice can remove probe-dependent uncertainty more accurately than an eavesdropper, Eve, because Alice knows the transmitted probes exactly while Eve only has noisy observations of them. Across the main STEEP formulations, the resulting secrecy rate is positive whenever Eve’s probe observation is not noiseless and the echoing phase is provisioned appropriately, including regimes in which Eve’s receive channels are stronger than those of the legitimate users in both directions (Hua, 2023, Hua, 2024, Hua, 2024, Hua, 7 Aug 2025).

1. Origin and conceptual setting

STEEP was introduced as a two-phase scheme motivated by the Maurer, Ahlswede and Csiszár bounds on secret-key capacity for channel probing over single-input and single-output channels (Hua, 2023). The original formulation emphasizes that conventional wiretap secrecy capacity for one-way communication is zero when the eavesdropper’s channel is as strong as or stronger than the main channel, whereas channel probing with public communication can still support secret-key generation if Eve’s observation is noisy (Hua, 2023).

The basic STEEP construction adapts that insight from key generation to direct secret-message transmission. In phase 1, Alice sends random probes over a probing channel to Bob. In phase 2, Bob echoes an estimated version of those probes, but encrypted by a secret, over a return channel (Hua, 2023). This induces an effective wiretap channel from Bob to Alice and Eve, with the intended asymmetry arising not from a better physical return channel alone, but from Alice’s exact probe knowledge relative to Eve’s imperfect estimate (Hua, 2023, Hua, 2024).

Subsequent work generalized the scheme from SISO settings to MIMO Gaussian channels, PSK-based nonlinear constructions, and multiple-access scenarios, while also framing STEEP as a unification of secret-key generation and wiretap-channel transmission (Hua, 2024, Hua et al., 2024). A later treatment presented STEEP explicitly as an alternative to quantum key distribution (QKD), emphasizing classical-channel operation, undersea optical links, and secrecy rates sufficient for one-time-pad encryption in many practical situations (Hua, 7 Aug 2025).

2. Protocol architecture and operating assumptions

The standard STEEP protocol is organized into two non-overlapping phases. In the probing phase, Alice transmits random probe symbols to Bob; Bob and Eve each observe noisy versions of those probes (Hua, 2023, Hua, 7 Aug 2025). In the echoing phase, Bob transmits a signal formed from his probe observation and the secret message, using either linear or nonlinear encryption depending on the variant (Hua, 2024, Hua, 7 Aug 2025).

For the SISO AWGN formulation described in the 2025 treatment, phase 1 is

$$y_{1,k} = x_{1,k} + w_{1,k}, \qquad z_{1,k} = x_{1,k} + v_{1,k},$$

where $x_{1,k}$ is Alice’s random probe, $w_{1,k}$ is AWGN with variance $\sigma_1^2$, and $v_{1,k}$ is AWGN with variance $\epsilon_1^2$ (Hua, 7 Aug 2025). Eve’s channel advantage in phase 1 is defined as

$\alpha_1 = \sigma_1^2 / \epsilon_1^2$

(Hua, 7 Aug 2025).

In phase 2, Bob constructs

$x_{2,k} = c_1 y_{1,k} + c_2 s_{2,k},$

with non-negative real weights $c_1, c_2$, where $s_{2,k}$ is the secret message (Hua, 7 Aug 2025). Alice and Eve receive

$x_{1,k}$0

with $x_{1,k}$1 and $x_{1,k}$2 AWGN of variances $x_{1,k}$3 and $x_{1,k}$4, respectively, and $x_{1,k}$5 quantifying Eve’s phase-2 advantage (Hua, 7 Aug 2025).

The broader literature stresses several operating assumptions. STEEP does not require full-duplex operation, channel reciprocity, secure feedback channel, collaborative third party, or Eve’s channel state information at the legitimate users (Hua, 2024, Hua et al., 2024). Eve may have any given number of antennas and may know all channel states (Hua, 2024). The essential requirement is that Eve’s receive channel in the probing phase is not noiseless (Hua, 2024, Hua et al., 2024). In the 2025 presentation, all parties are assumed to have access to conventional coding and hashing tools, and regenerative relays are identified as problematic while non-regenerative repeaters are compatible with the scheme (Hua, 7 Aug 2025).

3. Information-theoretic mechanism

The security mechanism is probe-dependent asymmetry. Alice can remove the probe contribution from Bob’s echoed transmission because she knows the original probes exactly, while Eve can only subtract an estimate based on noisy probe observations (Hua, 2023, Hua, 2024). This creates a favorable main channel for Alice over the return link even when Eve had stronger channels during probing and echo reception (Hua, 2023, Hua, 2024).

In the early SISO analog formulation, the secrecy rate per probe sample achieved by STEEP is

$x_{1,k}$6

where $x_{1,k}$7 and $x_{1,k}$8 (Hua, 2023). This expression is positive for any finite $x_{1,k}$9, even if $w_{1,k}$0 (Hua, 2023).

The same paper describes the induced return-link wiretap channel through

$w_{1,k}$1

for Alice, and

$w_{1,k}$2

for Eve, where $w_{1,k}$3 is Eve’s probe estimation error (Hua, 2023). Alice’s effective SNR is therefore higher than Eve’s because Eve’s residual probe uncertainty enters her decoder explicitly (Hua, 2023).

The 2024 revisiting paper makes this dependence on probe noise and echo design more explicit. It states that STEEP yields a positive secrecy rate in bits per channel use even if the receive channels at Eve are stronger than those between legitimate users in both forward and reverse directions, provided the power in the echoing phase is sufficiently large and Eve’s receive channel in the probing phase is not noiseless (Hua, 2024). It further states that, under asymmetric large powers in forward and reverse directions, the secrecy rate of G-STEEP approaches the secret-key capacity based on Gaussian probing signal over MIMO Gaussian channel (Hua, 2024).

A plausible implication is that STEEP should be understood less as a conventional degraded-wiretap construction and more as a round-trip synthesis of a new effective wiretap channel whose advantage derives from probe knowledge rather than raw link superiority. That interpretation is consistent with the stated relation to Maurer’s SKG protocol and Hayashi’s two-way schemes (Hua, 2024).

4. Formal secrecy-rate expressions and asymptotics

The 2025 AWGN treatment uses the secrecy rate

$w_{1,k}$4

where $w_{1,k}$5 is the conditional mutual information available to the legitimate receiver about $w_{1,k}$6, and $w_{1,k}$7 is the corresponding quantity for Eve (Hua, 7 Aug 2025). For Alice,

$w_{1,k}$8

and her MMSE is

$w_{1,k}$9

(Hua, 7 Aug 2025). For Eve,

$\sigma_1^2$0

with

$\sigma_1^2$1

(Hua, 7 Aug 2025).

The per-use secrecy rate is then given as

$\sigma_1^2$2

subject to

$\sigma_1^2$3

and $\sigma_1^2$4, $\sigma_1^2$5, where $\sigma_1^2$6 and $\sigma_1^2$7 (Hua, 7 Aug 2025). The paper states that $\sigma_1^2$8 is achievable even if $\sigma_1^2$9, as long as transmission power in phase 2 is sufficiently high relative to phase 1 and Eve’s channel advantage (Hua, 7 Aug 2025). In the high-SNR regime,

$v_{1,k}$0

which remains strictly positive as long as Eve is not perfectly co-located (Hua, 7 Aug 2025).

The 2024 unification paper gives the SISO secret-key capacity as

$v_{1,k}$1

which is always positive as long as $v_{1,k}$2 and $v_{1,k}$3 (Hua et al., 2024). For the MISO setting it gives

$v_{1,k}$4

where $v_{1,k}$5 and $v_{1,k}$6 are effective noise variances after Alice’s probe cancellation and after Eve’s estimation error is accounted for (Hua et al., 2024). The same paper states that if $v_{1,k}$7 and $v_{1,k}$8 is large, $v_{1,k}$9 scales like $\epsilon_1^2$0, while if $\epsilon_1^2$1, $\epsilon_1^2$2 remains positive but saturates at high power (Hua et al., 2024).

For Gaussian MIMO channels, the revisiting paper defines

$\epsilon_1^2$3

with determinant expressions involving the relevant covariance and MMSE matrices, and gives the secret-key-capacity limit

$\epsilon_1^2$4

(Hua, 2024). It also states the high-SNR degree of freedom

$\epsilon_1^2$5

(Hua, 2024).

5. Variants and extensions

The STEEP literature distinguishes several variants.

Variant	Channel model	Core construction
G-STEEP	MIMO Gaussian	Gaussian probing and Gaussian linear encryption
P-STEEP	SISO PSK	PSK probing and nonlinear encryption
M-STEEP	Multiple access	Shared Gaussian probing and orthogonal echoing

These variants are explicitly identified in the revisiting paper (Hua, 2024).

For G-STEEP, Alice sends Gaussian random vectors $\epsilon_1^2$6, Bob forms an MMSE estimate of the effective probe, and Bob transmits an encrypted combination of that estimate and a Gaussian secret message $\epsilon_1^2$7 (Hua, 2024). Positive secrecy rate is achievable for any channel and any number of antennas at Eve whenever Bob’s echoing power is large enough and Eve’s probe channel is not noiseless (Hua, 2024).

For P-STEEP, the probe is $\epsilon_1^2$8 with $\epsilon_1^2$9 uniformly over $\alpha_1 = \sigma_1^2 / \epsilon_1^2$0 phases, and Bob transmits $\alpha_1 = \sigma_1^2 / \epsilon_1^2$1, where $\alpha_1 = \sigma_1^2 / \epsilon_1^2$2 is a soft estimate and $\alpha_1 = \sigma_1^2 / \epsilon_1^2$3 carries the secret phase (Hua, 2024). The paper gives

$\alpha_1 = \sigma_1^2 / \epsilon_1^2$4

$\alpha_1 = \sigma_1^2 / \epsilon_1^2$5

and

$\alpha_1 = \sigma_1^2 / \epsilon_1^2$6

with positivity condition $\alpha_1 = \sigma_1^2 / \epsilon_1^2$7, assured if

$\alpha_1 = \sigma_1^2 / \epsilon_1^2$8

(Hua, 2024).

For M-STEEP, an access point with multiple antennas broadcasts Gaussian probes shared among multiple users, and each user echoes an encrypted probe estimate with its message using orthogonal multiple access (Hua, 2024). For user 1,

$\alpha_1 = \sigma_1^2 / \epsilon_1^2$9

and each user’s secrecy rate is positive if that user’s echoing power is sufficiently large (Hua, 2024).

The 2023 paper additionally describes a digital or upper-layer adaptation in which Alice transmits random bitstreams, Bob later XORs a random sequence with his version of those bits, and the secrecy rate per bit is

$x_{2,k} = c_1 y_{1,k} + c_2 s_{2,k},$0

where $x_{2,k} = c_1 y_{1,k} + c_2 s_{2,k},$1 (Hua, 2023). This suggests that the protocol family is not restricted to analog PHY-layer probing but can be abstracted to noisy observation asymmetries in connected networks more generally (Hua, 2023).

6. Relation to secret-key generation, feedback, and prior information-theoretic schemes

A recurring theme in the STEEP literature is its connection to secret-key generation. The original STEEP paper explicitly derives motivation from the Maurer, Ahlswede and Csiszár bounds and states that, for one-way SISO channel probing, the MAC lower and upper bounds coincide (Hua, 2023). The revisiting paper states that G-STEEP, with asymmetric large powers in forward and reverse directions, has its secrecy rate approaching the secret-key capacity based on Gaussian probing signal over MIMO Gaussian channel (Hua, 2024). The unification paper makes the connection explicit in its title and argues that STEEP yields a positive secrecy rate in every channel coherence period and that this rate does not diminish as coherence time increases (Hua et al., 2024).

The role of public or observable feedback is also central in related information-theoretic secrecy results. In the broadcast erasure-channel setting with public strictly causal state-feedback, a two-phase scheme first creates appropriate secret keys and then uses them to encrypt each message, and the amount of key needed is smaller than the size of the message and equal to the amount of encrypted message the potential eavesdroppers jointly collect (Czap et al., 2014). That work characterizes secure communication over a 1-to-$x_{2,k} = c_1 y_{1,k} + c_2 s_{2,k},$2 broadcast erasure channel with public state-feedback and shows that feedback enables key generation by exploiting packets received by only one receiver (Czap et al., 2014).

The relation to STEEP is not identity of model but analogy of architecture. Both frameworks use a two-phase design, exploit asymmetries created by stochastic observation differences, and reduce secrecy overhead by aligning it with actual leakage rather than total message size (Czap et al., 2014). This suggests that STEEP belongs to a broader class of feedback-enabled or round-trip physical-layer secrecy constructions in which randomness first creates correlated but unequal observations and is then converted into secrecy by carefully structured transmission.

The same 2014 erasure-channel result also proves that a dishonest receiver that provides deceptive feedback cannot diminish the rate experienced by the honest receivers (Czap et al., 2014). A plausible implication is that robustness to protocol-visible side information and asymmetry management are already present in adjacent feedback-security literatures, even though STEEP itself is framed for probe-and-echo channels rather than broadcast erasures.

7. Practical significance, comparisons, and limitations

The 2025 paper positions STEEP as an alternative to QKD, stating that it requires only classical or non-quantum channels and can operate over existing communication infrastructures such as in-air channels or optical cables (Hua, 7 Aug 2025). It further states that STEEP is compatible with non-regenerative repeaters, which match modern undersea optical links, whereas regenerative relays can break the secrecy property (Hua, 7 Aug 2025). The same source states that STEEP can yield a secrecy rate sufficient for one-time pads encryption in many practical situations and reports minimum achievable secrecy rates such as $x_{2,k} = c_1 y_{1,k} + c_2 s_{2,k},$3 bits/use in worst-case adversarial positioning for typical undersea fiber cable parameters, as per Figure 1 (Hua, 7 Aug 2025).

The comparative claims against QKD in that treatment concern hardware and deployment assumptions rather than a universal dominance theorem. STEEP is described as requiring no quantum hardware, no dual channel, and no quantum repeaters, and as using off-the-shelf classical FEC and coding hardware (Hua, 7 Aug 2025). It is also described as robust against constant eavesdropping, in contrast to the statement that constant intercept can disable QKD key generation (Hua, 7 Aug 2025). These are claims about the model and scenarios considered in that paper, not a general equivalence between the security frameworks.

Several limitations are explicit in the literature. First, STEEP’s positivity claims require that Eve’s probe channel not be noiseless (Hua, 2023, Hua, 2024, Hua et al., 2024). Second, positive secrecy commonly depends on sufficiently large echoing power or asymmetric power allocation (Hua, 2024, Hua, 7 Aug 2025). Third, in some implementations, authentication on the return channel is assumed to prevent active manipulation (Hua, 2023). Fourth, specific deployment feasibility can depend on the analog path being non-regenerative (Hua, 7 Aug 2025).

A common misconception is that STEEP requires channel reciprocity or full-duplex hardware because it is a round-trip protocol. The papers explicitly reject both requirements: the protocol operates with half-duplex phases and arbitrary, potentially asymmetric, channels (Hua, 2024, Hua et al., 2024). Another misconception is that a stronger Eve necessarily forces zero secrecy rate. The published STEEP results state the opposite under their assumptions: secrecy can remain positive even when Eve is stronger in both directions, provided her probe observation is not noiseless and the echoing phase is designed appropriately (Hua, 2024, Hua, 7 Aug 2025).

Taken together, the STEEP literature presents a family of classical-channel, round-trip secrecy schemes whose core invariant is probe asymmetry. Its contribution lies not in eliminating the need for physical asymmetry, but in relocating that asymmetry from raw channel superiority to differential knowledge of the probe, then amplifying it through echo design, coding, and power allocation (Hua, 2023, Hua, 2024, Hua, 7 Aug 2025).