Poster Summary: Towards Large-Scale Economic-Robust Spectrum Auctions

Dynamic spectrum auction is an effective solution to manage spectrum across many small networks. Yet without proper economic design, spectrum auctions can be easily manipulated by bidders, suffering huge efficiency loss. Selfish bidders can manipulate their bids, individually or in groups, to game the system and obtain outcomes that favor their own interests and hurt others. Prior solutions have designed truthful auctions to tackle individual bidder cheating [8]. As the number of participants grows, collusion becomes a serious threat to auction revenue and efficiency. Extensive measurements [2], [3] have shown that in many past auctions including the FCC spectrum auctions, a small fraction (<5%) of bidders have strategically formed one or multiple collusion groups and rig their bids to manipulate auction results, causing lower prices and unfair resource distribution. Because collusion is legally banned in commercial auctions, existing collusion groups were tacit and small in size, thus easier to form and hard (and expensive) to detect in large-scale auctions. Similar trends were observed in other practical deployments including P2P systems where each collusion group contains 2–4 players [5]. In this work, we show that in spectrum auctions, small-size collusion is even more effective than that in conventional auctions, because colluding bidders can exploit the bidder interference constraints to rig their bids. On the other hand, designed to addressing individual cheating, truthful auctions are highly vulnerable to collusion. These observations pose an important need for spectrum auctions to resist bidder collusion, particularly small-size collusion. Prior solutions to tackle collusion in auctions, however, when applied to spectrum auctions, either cause severe interference or lose collusion-resistance [9]. This is because conventional designs [4], [6] do not consider any reuse and assume bidders have a homogeneous relationship: either all conflict with each other, or do not conflict at all. Whereas this relationship in spectrum auctions becomes heterogeneous due to the bidder interference constraints and the need for spatial reuse [8]. On the other hand, recent work on spectrum auctions focuses on suppressing some forms of collusion [7], but can be attacked easily by other simple forms. More importantly, this solution requires an exponential-complexity algorithm to ensure its resistance, thus cannot operate in large-scale dynamic spectrum auctions. In this work, we propose DC, a new collusion-resistant and computationally-efficient spectrum auction. Using a randomization technique, DC resists collusion by diminishing the gain of any colluding group unless it becomes large (and hence hard to form and easy to be detected). Such diminishing returns leave bidders little or no incentive to collude. Meanwhile, DC enables spatial reuse to improve auction revenue and efficiency. DC’s novel contribution is to let the auctioneer secretely perform a 3-stage “Divide, Conquer, and Combine” procedure after receiving bids. By judiciously designing its procedure, DC successfully integrates an efficient spectrum allocation algorithm (in “Divide”) with a novel economic mechanism (in “Conquer”), enabling spatial reuse and effectively controlling collusion.