---
title: 'STEEP: Echoing Encrypted Probes for Secure Messaging'
url: https://www.emergentmind.com/topics/secret-message-transmission-by-echoing-encrypted-probes-steep
type: topic
---

# STEEP: Echoing Encrypted Probes for Secure Messaging

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 [2309.14529][2410.03515][2410.03515][2508.05882].

## 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 [2309.14529]. 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 [2309.14529].

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 [2309.14529]. 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 [2309.14529][2410.03515].

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 [2410.03515][2403.06438]. 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 [2508.05882].

## 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 [2309.14529][2508.05882]. 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 [2410.03515][2508.05882].

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 \(\epsilon_1^2\) [2508.05882]. Eve’s channel advantage in phase 1 is defined as
\[
\alpha_1 = \sigma_1^2 / \epsilon_1^2
\]
[2508.05882].

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 [2508.05882]. Alice and Eve receive
\[
y_{2,k} = x_{2,k} + w_{2,k},
\qquad
z_{2,k} = x_{2,k} + v_{2,k},
\]
with \(w_{2,k}\) and \(v_{2,k}\) AWGN of variances \(\sigma_2^2\) and \(\epsilon_2^2\), respectively, and \(\alpha_2 = \sigma_2^2 / \epsilon_2^2\) quantifying Eve’s phase-2 advantage [2508.05882].

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 [2410.03515][2403.06438]. Eve may have any given number of antennas and may know all channel states [2410.03515]. The essential requirement is that Eve’s receive channel in the probing phase is not noiseless [2410.03515][2403.06438]. 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 [2508.05882].

## 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 [2309.14529][2410.03515]. This creates a favorable main channel for Alice over the return link even when Eve had stronger channels during probing and echo reception [2309.14529][2410.03515].

In the early SISO analog formulation, the secrecy rate per probe sample achieved by STEEP is
\[
\boxed{
\xi_{STEEP,AC}
=
\mathbb{E} \left\{
\log \left( 1 + \frac{SNR_{B,A}}{1 + SNR_{E,A}} \right)
\right\}
}
\]
where \(SNR_{B,A} = \frac{p_A |h_{B,A}|^2}{\sigma_B^2}\) and \(SNR_{E,A} = \frac{p_A \|\mathbf{g}_A\|^2}{\sigma_{E,A}^2}\) [2309.14529]. This expression is positive for any finite \(SNR_{E,A}\), even if \(SNR_{E,A} > SNR_{B,A}\) [2309.14529].

The same paper describes the induced return-link wiretap channel through
\[
t_A(k) = y_{A,B}(k) - h_{B,A} x_A(k) = s(k) + w_B(k) + v_A(k),
\]
for Alice, and
\[
t_E(k) = y_{E,B}(k) - h_{B,A} \hat{x}_A(k)
= s(k) + h_{B,A} \Delta x_A(k) + w_B(k) + v_E(k),
\]
for Eve, where \(\Delta x_A(k)\) is Eve’s probe estimation error [2309.14529]. Alice’s effective SNR is therefore higher than Eve’s because Eve’s residual probe uncertainty enters her decoder explicitly [2309.14529].

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 [2410.03515]. 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 [2410.03515].

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 [2410.03515].

## 4. Formal secrecy-rate expressions and asymptotics

The 2025 AWGN treatment uses the secrecy rate
\[
R_s = \max(0, C_U - C_E),
\]
where \(C_U\) is the conditional mutual information available to the legitimate receiver about \(s_{2,k}\), and \(C_E\) is the corresponding quantity for Eve [2508.05882]. For Alice,
\[
C_U = I(s_{2,k}; y_{2,k}, x_{1,k}) = I(s_{2,k}; y_{2,k}|x_{1,k}),
\]
and her MMSE is
\[
MSE_U = \frac{1}{1 + \frac{c_2^2}{c_1^2 \sigma_1^2 + \sigma_2^2}}
\]
[2508.05882]. For Eve,
\[
C_E = I(s_{2,k}; z_{2,k}, z_{1,k}),
\]
with
\[
MSE_E =
\frac{1}{1 + \frac{c_2^2 (1+\epsilon_1^2)}{(1 + \epsilon_1^2)(1+\epsilon_2^2 - c_2^2) - c_1^2}}
\]
[2508.05882].

The per-use secrecy rate is then given as
\[
R_s =
\log_2
\frac{
1 + \frac{c_2^2}{c_1^2 \sigma_1^2 + \sigma_2^2}
}{
1 + \frac{c_2^2 (1+\epsilon_1^2)}{(1 + \epsilon_1^2)(1+\epsilon_2^2 - c_2^2) - c_1^2}
}
\]
subject to
\[
c_1^2 (1 + \sigma_1^2) + c_2^2 = 1
\]
and \(\epsilon_1^2 = 1/(\alpha_1 p_1)\), \(\epsilon_2^2 = 1/(\alpha_2 p_2)\), where \(p_1 = 1/\sigma_1^2\) and \(p_2 = 1/\sigma_2^2\) [2508.05882]. The paper states that \(R_s > 0\) is achievable even if \(\alpha_1, \alpha_2 > 1\), as long as transmission power in phase 2 is sufficiently high relative to phase 1 and Eve’s channel advantage [2508.05882]. In the high-SNR regime,
\[
\lim_{p_1\to\infty, p_2\to\infty} R_s = \log_2 (1 + 1/\alpha_1),
\]
which remains strictly positive as long as Eve is not perfectly co-located [2508.05882].

The 2024 unification paper gives the SISO secret-key capacity as
\[
C_{key} = \mathbb{E}\left\{ \log \left( 1 + \frac{\mathrm{SNR}_m}{1 + \mathrm{SNR}_e} \right) \right\},
\]
which is always positive as long as \(\mathrm{SNR}_m > 0\) and \(\mathrm{SNR}_e < \infty\) [2403.06438]. For the MISO setting it gives
\[
C_{STEEP}
=
\log \left( 1 + \frac{1}{\sigma_{v,A}^2} \right)
-
\log \left( 1 + \frac{1}{\sigma_{v,E}^2} \right),
\]
where \(\sigma_{v,A}^2\) and \(\sigma_{v,E}^2\) are effective noise variances after Alice’s probe cancellation and after Eve’s estimation error is accounted for [2403.06438]. The same paper states that if \(n_A > n_E\) and \(P_A\) is large, \(C_{STEEP}\) scales like \(\log P_A\), while if \(n_A \leq n_E\), \(C_{STEEP}\) remains positive but saturates at high power [2403.06438].

For Gaussian MIMO channels, the revisiting paper defines
\[
C_{s,G} = \left[ C_{A|B,G} - C_{E|B,G} \right]^+
\]
with determinant expressions involving the relevant covariance and MMSE matrices, and gives the secret-key-capacity limit
\[
C_{key} =
\log \left|\mathbf{I} + \mathbf{H}_{BA}'^H \mathbf{H}_{BA}' (\mathbf{H}_{EA}'^H \mathbf{H}_{EA}' + \mathbf{I})^{-1} \right|
\]
[2410.03515]. It also states the high-SNR degree of freedom
\[
\text{DoF}(C_{s,G}) = \min(n_B, (n_A-n_E)^+)
\]
[2410.03515].

## 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 [2410.03515].

For G-STEEP, Alice sends Gaussian random vectors \(\mathbf{x}_A \sim \mathcal{CN}(0,\mathbf{I})\), Bob forms an MMSE estimate of the effective probe, and Bob transmits an encrypted combination of that estimate and a Gaussian secret message \(\mathbf{s} \sim \mathcal{CN}(0,\mathbf{I})\) [2410.03515]. 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 [2410.03515].

For P-STEEP, the probe is \(x_A = e^{j\theta}\) with \(\theta\) uniformly over \(M\) phases, and Bob transmits \(x_B = e^{j\phi} r_B\), where \(r_B\) is a soft estimate and \(\phi\) carries the secret phase [2410.03515]. The paper gives
\[
p_{e,A} = n_0 Q\left( \frac{ \sin(\pi/M) }{ \sqrt{ 1/(2a) + 1/(2b) } } \right),
\]
\[
p_{e,E} = n_0 Q\left( \frac{ \sin(\pi/M) }{ \sqrt{ 1/(2a) + 1/(2 S_{EA}) + 1/(2 S_{EB}) } } \right),
\]
and
\[
C_{s,P} = [ h_2(p_{e,E}) - h_2(p_{e,A}) ]^+,
\]
with positivity condition \(p_{e,A} < p_{e,E}\), assured if
\[
\frac{b}{a} > \alpha \left(1 - \frac{1}{\beta}\right)
\]
[2410.03515].

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 [2410.03515]. For user 1,
\[
C_{s,1} =
\left[
\log \left( 1 + \frac{ S_{A,1}/2 }{ \frac{S_1 S_{A,1}/2}{(S_1+1)^2} + 1 } \right)
-
\log \left( 1 + \frac{ S_{E,1}/2 }{ (\gamma_1 - 1 ) S_{E,1}/2 + 1 } \right)
\right]^+
\]
and each user’s secrecy rate is positive if that user’s echoing power is sufficiently large [2410.03515].

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
\[
\xi_{STEEP,DC} = f(P_{E|B}) - f(P_{A|B}),
\]
where \(f(p) = -p\log p - (1-p)\log(1-p)\) [2309.14529]. 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 [2309.14529].

## 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 [2309.14529]. 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 [2410.03515]. 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 [2403.06438].

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 [1408.1800]. That work characterizes secure communication over a 1-to-\(K\) broadcast erasure channel with public state-feedback and shows that feedback enables key generation by exploiting packets received by only one receiver [1408.1800].

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 [1408.1800]. 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 [1408.1800]. 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 [2508.05882]. 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 [2508.05882]. 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 \(0.16\) bits/use in worst-case adversarial positioning for typical undersea fiber cable parameters, as per Figure 6 [2508.05882].

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 [2508.05882]. It is also described as robust against constant eavesdropping, in contrast to the statement that constant intercept can disable QKD key generation [2508.05882]. 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 [2309.14529][2410.03515][2403.06438]. Second, positive secrecy commonly depends on sufficiently large echoing power or asymmetric power allocation [2410.03515][2508.05882]. Third, in some implementations, authentication on the return channel is assumed to prevent active manipulation [2309.14529]. Fourth, specific deployment feasibility can depend on the analog path being non-regenerative [2508.05882].

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 [2410.03515][2403.06438]. 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 [2410.03515][2508.05882].

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 [2309.14529][2410.03515][2508.05882].

Source: https://www.emergentmind.com/topics/secret-message-transmission-by-echoing-encrypted-probes-steep