Abstract-We present necessary and sufficient conditions for stochastic stabilizability of unstable (non-stationary) noisy linear systems over channels with memory with feedback. Stochastic stability notions include recurrence, Birkhoff sample path ergodicity and asymptotic mean stationarity, and the existence of finite second moments. We extend recent results in the literature on noiseless and erasure channels for systems driven by possibly unbounded noise. Our constructive proof uses randomtime state-dependent stochastic drift criteria for stabilization of Markov chains. For asymptotic mean stationarity, it is sufficient that the capacity of a channel is (strictly) greater than the sum of the logarithms of the unstable pole magnitudes for memoryless channels and a class of channels with memory. We also present the tightness of the sufficient condition under a technical condition. We provide sufficiency conditions for the existence of finite average second moments, for such systems driven by unbounded noise, which had not been studied in the literature to our knowledge. Comparison with relevant results in the literature is presented.
I. PROBLEM FORMULATIONWe consider a scalar LTI discrete-time system described bywhere x t is the state at time t, u t is the control input, the initial condition x 0 is a second order random variable, and {d t } is a sequence of zero-mean independent, identically distributed (i.i.d.) Gaussian random variables with a finite second moment. It is assumed that |a| ≥ 1 and b = 0:The system is open-loop unstable, but it is stabilizable. This system is connected over a Discrete Memoryless Channel with a finite capacity to a controller, as shown in Figure 1.The channel source consists of state values, taking values in R. The source is quantized: A quantizer Q is represented by a map Q : R → R, characterized by a sequence of nonoverlapping Borel-measurable bins {B i }, such that Q(x) = q i ∈ R if and only if x ∈ B i ; that is, Q(x) = i q i 1 {x∈Bi} . The quantizer outputs are transmitted through a channel, after being subjected to a channel encoder. Definition 1.1: A finite-alphabet channel with memory is characterized by a sequence of finite input alphabets M n+1 , finite output alphabets M ′ n+1 , and a sequence of conditional probability measures P n (q ′ , and a conditional probability mass functionsatisfies the following:The quantizer outputs are transmitted through a noisy channel, hence the receiver has access to noisy versions of the quantizer/coder outputs for each time, which we denote by q ′ ∈ M ′ . The quantizers and controllers are causal such that the quantizer function at time t ≥ 0 is generated using the information vector I The goal of the paper is to identify conditions on the channel under which the controlled process {x t } is stochastically stable in each of the following senses:• {x t } is recurrent: There exists a compact set A such that {x t ∈ A} infinitely often almost surely.• {x t } is asymptotically mean stationary and satisfies the sample path ergodic theorem.2 e...