论文解读

AI 辅助整理
方法
对比Pectra与Fusaka升级前后ECDSA、BLS等签名及批量、Merkl
发现
批量与Merkle锚定使ECDSA成本降六倍至每条8963 gas,单笔可承载1754条记录。
机会点
尚未实地部署,可开展真实农业供应链场景的链上认证成本与吞吐验证研究。

论文摘要

研究农业供应链金融IoT数据的区块链认证成本,对比Pectra与Fusaka升级前后四种签名方案。批量与Merkle锚定使ECDSA成本降六倍至每条8963 gas;单笔最多承载1754条ECDSA记录,BLS仅411条。未实地部署。

全文

arXiv

Authenticating IoT Data on Ethereum after Pectra and Fusaka:
A Cost Study of Signature Schemes, Batching and Anchoring
for Agricultural Supply Chain Finance
Ming-Fong Siea, Jian-Hong Chenb, Gwan-yen Linc, Yen-Shuo Chenc, Shih-wei Liaob,*
aMaster Program in Intelligent Computing and Big Data, Chung Yuan Christian University, Taoyuan, Taiwan
bDepartment of Computer Science and Information Engineering, National Taiwan University, Taipei, Taiwan
cGraduate Institute of Networking and Multimedia, National Taiwan University, Taipei, Taiwan
*Corresponding author. E-mail: liao@csie.ntu.edu.tw
Abstract
Blockchain-based supply chain finance can turn the sensor-recorded production history of small-
holder farms into verifiable credit evidence, but only if every record is authenticated and anchored at
a cost that scales to large device fleets. Ethereum’s 2025 Pectra and Fusaka upgrades changed this
cost landscape: they added precompiles for BLS12-381 and P-256, repriced calldata and MODEXP,
and capped transaction gas at 224. We organize the authentication and anchoring of IoT records into
three trust levels, from posting a Merkle root through a trusted aggregator to verifying every device
signature on-chain, and implement four replay-protected signature designs: ECDSA over secp256k1
and P-256, BLS aggregate signatures with on-chain hash-to-curve, and a pseudonymous identity-
based signature with batch verification. Measured as signed transactions under the rules before and
after both upgrades, the upgrades mainly changed which designs are feasible, not their cost. Batch-
ing and Merkle anchoring cut the cost of ECDSA authentication sixfold, to 8,963 gas per record, and
secp256k1 ECDSA remains the cheapest authenticated design: P-256 costs 1.6 times as much, the
identity-based design 2.7 times and BLS aggregation 3.9 times, dominated by on-chain hashing to
the curve. One transaction carries at most 1,754 ECDSA-authenticated records but only 411 BLS-
aggregated ones. On a Raspberry Pi Zero 2 W, pairing-based signing takes 5.7 ms against 0.29 ms for
ECDSA. Pairing-based designs pay off only when the application needs their additional properties,
and the identity-based design adds key escrow at the key generation centre. The benchmark harness
is released for reproducibility.
Keywords:IoT data authentication; Ethereum gas; batch verification; pairing-based signatures; ECDSA;
supply chain finance
1 Introduction
Smallholder farmers are the most upstream and the most financially fragile participants in agricultural
supply chains. They rarely keep auditable records, hold little collateral and therefore struggle to obtain
credit [1]. Blockchain-based supply chain systems make production and transaction data tamper-evident
and shareable among farmers, distributors and lenders [2–5], and verifiable transparency is known to
1
arXiv:2611.23764v1 [cs.CR] 18 Nov 2026

lower the cost of financing [6]. If the production history of a farm can be verified, lenders can release
staged payments when production milestones are reached.
The evidence comes from devices: in our deployment scenario, low-cost camera units photograph
crops at fixed intervals, store the images off-chain and upload signed fingerprints. Every record must
be authenticated, otherwise production histories can be fabricated, and the cost of doing so must scale
to thousands of records per day. Where should authentication happen, with which signature scheme,
and at what cost? Two recent Ethereum upgrades make these questions worth revisiting. Pectra (May
2025) [7] added precompiles for the BLS12-381 pairing-friendly curve (EIP-2537 [8]) and increased the
price of data-heavy calldata (EIP-7623 [9]). Fusaka (December 2025) [10, 11] added a P-256 signature-
verification precompile (EIP-7951 [12]), repriced MODEXP (EIP-7883 [13]), raised the default block
gas limit to 60 million (EIP-7935 [14]) and capped the gas of every transaction at 224 =16,777,216 (EIP-
7825 [15]). Pairing-based schemes that were impractical on-chain before are now possible, standard
device signatures can be verified natively, and a single transaction can no longer absorb an arbitrarily
large batch.
The literature on IoT blockchains rarely quantifies these trade-offs. Proposals typically adopt one
scheme, often a pairing-based identity-based signature with batch verification, and evaluate its crypto-
graphic operation counts off-chain, without comparing it to the simplest alternative of verifying standard
ECDSA signatures with the precompiles every Ethereum client provides. The Multi-Layer Aggregate
Verification (MLA V) design [16, 17], in which two of the authors were involved, is an example: it
used an identity-based batch-verification signature of the IBV family [18] and reported only simulated,
operation-count-based costs. Re-examining it, we found that the published signature is insecure, inherit-
ing weaknesses already known for this family [19], which motivated us to measure, rather than assume,
what each design costs.
Contributions.
1.Design space and trust levels.We organize the authentication and anchoring of IoT records into
three trust levels (Section 4): T0, a trusted aggregator posts a Merkle root; T1, it posts all record
hashes; T2, a contract verifies every device signature. Within T2 we consider individual submission
and batch submission with Merkle anchoring.
2.Four implemented designs.We implement ECDSA over secp256k1 (ecrecover) and P-256 (EIP-
7951), BLS aggregate signatures with an RFC 9380-conformant on-chain hash-to-curve built from
the EIP-2537 and MODEXP precompiles, and MLA V-IBS, a corrected pseudonymous identity-based
signature with small-exponent and pseudonym-grouped batch verification (Section 5).
3.Measurements under current rules.We execute all designs as signed transactions on an EVM con-
figured with the Osaka (Fusaka) rules, and under the Cancun and Prague rules that preceded them, and
report gas per record, the largest batch per transaction under EIP-7825, block-level capacity, calldata,
deployment and registration costs, and device- and aggregator-side timings (Sections 6 and 7).
4.Findings and guidelines.We quantify when batching, anchoring and each signature scheme pay off,
and derive guidelines for practitioners (Section 8). All code and data are released.
2

Relation to prior work.This article is a new study that builds on the MLA V architecture, introduced
in a two-page conference paper [16] and described in full in a journal article [17], both co-authored by
the first and last authors. It is not an extended version of either. The three-layer architecture (Figure 1)
and the supply chain finance workflow (Figure 2) are adapted from those works; both figures have been
redrawn and revised. The signature analysed in Section 5.4 is the one specified in [17], and our analysis
shows that it is insecure. The design-space framework, all four implementations, the MLA V-IBS scheme,
the measurements and the analysis are new, and the results reported here supersede the simulated cost
comparisons of [16, 17].
2 Background and related work
Blockchains for agri-food supply chains and finance.Surveys document distributed ledgers for trace-
ability, food safety and certification in agriculture [2, 3]. Early systems combined RFID or sensor data
with a ledger [20, 21]; AgriBlockIoT demonstrated an implementation on Ethereum and Hyperledger [4];
later work addresses interoperability between enterprise chains [5]. Chod et al. [6] show that verifi-
able supply chain transparency lowers the cost of signalling operational quality to lenders. Compara-
ble ledgers have been built for pharmaceutical supply chains [22]. These systems usually authenticate
records by the transaction signature of the submitting account and do not analyze the cost of authenticat-
ing high-volume sensor data.
Blockchains for IoT.Architectures by Dorri et al. [23] and Novo [24] offload work from constrained
devices to more capable nodes, as the aggregators in our setting do. Lee and Lee [25] and Hu et al. [26]
used blockchains to distribute IoT firmware; the latter introduced pairing-based batch verification in this
context. Rollups move execution off the base layer [27]; Merkle anchoring is a simpler form of the same
idea that commits to records without making them available.
Signature schemes and batch verification.ECDSA over secp256k1 is Ethereum’s native transaction
signature and can be verified in a contract with theecrecoverprecompile; P-256 is the curve supported
by most secure elements, TPMs and WebAuthn authenticators. BLS signatures [28] on pairing-friendly
curves such as BLS12-381 [29] can be aggregated into a single signature [30, 31], at the price of hashing
every message to the curve [32]. Identity-based signatures (IBS) [33–36] remove certificates by deriving
public keys from identities. Batch verification checks many signatures together [37–39]; without random
weights it can be fooled [40]. In vehicular networks, Zhang et al.’s IBV scheme [18] started a large family
of pseudonymous identity-based batch-verification schemes, and Lee and Lai [19] pointed out that it is
vulnerable to replay, lacks non-repudiation and accepts batches of invalid signatures. Hakuta et al. [41]
studied batches of only a few signatures.
Pairings on Ethereum.Ethereum has offered BN254 pairing precompiles since 2017 [42]; BN254
provides roughly 100 bits of security, whereas BLS12-381 targets about 128. EIP-2537 adds BLS12-381
addition, multi-scalar multiplication (MSM), pairing-check and map-to-curve precompiles [8]. To our
knowledge, no prior study has compared pairing-based and ECDSA-based authentication of IoT data on
these precompiles under the Fusaka transaction cap.
3

Layer 1 — Ethereum (public)
Layer 2 — consortium chain (GoQuorum)
Layer 3 — field data (off-chain)
Device registry
address / key commitment
Verifier & anchoring contract
signature checks, Merkle roots
Government
node
Smallholder
node
Aggregator node
batch verification, Merkle tree
Financial
institution node
Distributor
node
shared consortium ledger
Smartphone app
farmers without sensors
IoP sensor (camera unit)
signs records (M, σ )
Off-chain storage
images and metadata
KGC
consortium-
operated
signed records
image
record hash
posted per batch:
T0 Merkle root
T1 + all record hashes
T2 + device signatures
verify
milestones
register keys
keys / pseudonyms
data key management query T0–T2: trust levels (Section 4)
Figure 1: MLA V platform (adapted from [16, 17]). Layer 3 collects field data off-chain, Layer 2 is a
permissioned consortium chain whose aggregator verifies records in batches, and Layer 1 is Ethereum.
What the aggregator posts per batch depends on the trust level T0–T2 of Section 4; the KGC issues
device keys and registers them on Layer 1.
3 Application context
Figure 1 shows the MLA V platform, adapted from [16, 17]. Akey generation centre(KGC), operated
by the consortium (for example the lender together with a public agency), registers farmers and their
devices and can trace a device back to its owner in case of dispute.IoP sensors(Raspberry Pi based
camera units) photograph crops, store the images off-chain and sign a recordM= (CT,T S,Chk,Seed)of
camera type, timestamp, image checksum and storage pointer (116 bytes in our encoding). Smallholders
without sensors use a smartphone application.Aggregator nodesof a permissioned consortium chain
(GoQuorum [43]) collect and verify records and anchor them on Ethereum.Financial institutionsgrant
loans and release staged payments, anddistribution channelssign purchase contracts and settle payments.
Figure 2 shows the financing workflow. A distributor and a farmer sign a purchase contract; the
farmer applies for a loan, and the lender approves it based on the contract and the farmer’s verified
production history, possibly with a deposit from the distributor. At each production checkpoint the
farmer’s devices upload signed records, which are verified and anchored, and the lender releases the
payment for that stage. If the agreed yield is not met, the lender recovers the outstanding instalments
and returns the deposit; otherwise the distributor pays the balance and the account is settled. Every step
leaves a verifiable record that becomes part of the farmer’s credit history.
4 Design space
We distinguish three trust levels, according to what a verifier who trusts only Ethereum can check.
•T0 – root only.The aggregator verifies records off-chain, builds a Merkle tree [44] and posts the root.
Anyone holding a record and its Merkle path can prove that the record was anchored at a given time,
4

Distributor Financial institution Smallholderwith IoP sensors Platformconsortium chain + Ethereum
register devices (KGC)
authenticate,
issue keys
sign purchase contract (smart contract)
anchor contract
Contract
stage
apply for loan
check contract and verified history
approve loan
deposit
Loan
stage
signed records of checkpoint i
verify and
anchor (T0–T2)
check milestone i
release payment for stage i
loop [each checkpoint i]
Production
stage
pay balance
settle account (net of loan)
repay instalments
return deposit
alt
[yield met]
[yield not met]
anchor settlement record (credit history)
Settle-
ment
message / data query payment
Figure 2: Supply chain finance workflow of MLA V (adapted from [17]), revised for this article. Signed
production records are verified and anchored at each checkpoint before the staged payment is released;
the settlement record becomes part of the farmer’s credit history.
but the chain neither reveals which records exist nor attests that they were signed by registered devices.
The aggregator is fully trusted for authenticity.
•T1 – record hashes.The aggregator posts all record hashes; the contract computes and stores the root.
The set of anchored records becomes public and cannot later be extended or changed, which makes
omissions and insertions detectable. Authenticity is still trusted to the aggregator.
•T2 – on-chain authentication.The contract verifies a device signature for every record before an-
choring it, so that anchored records are authenticated even if the aggregator is malicious. Records may
be submittedindividually, one transaction per record that stores the record hash, or inbatches, one
transaction per batch that verifies all signatures and stores a Merkle root.
All T2 designs share a device registry maintained by the KGC: a contract storing, per device, either its
address (secp256k1) or a 32-byte commitment to its public key, which the submitter provides in calldata.
Without such a registry, any key could sign records; with it, identity-based signatures lose their main
advantage of not requiring public-key distribution, a point we return to in Section 8.
5

Domain separation and replay protection.In every T2 design the device signs the record together
with a domaind= (chainid,contract), so a signature is valid only for the chain and the contract it was
made for. Signature verification alone does not stop a malicious aggregator from anchoring an old, validly
signed record again, for instance an old photograph presented as evidence for a new payment milestone,
and a batch is anchored as a single Merkle root, so the contract cannot recognise records it has already
anchored. Each design therefore stores, per device, the timestamp of its last accepted record and accepts
a record only if its timestamp is strictly larger. The slot is initialised at registration, so the same check
also rejects unregistered devices, and each device can be registered only once, so the KGC cannot reset
a device’s timestamp or replace its key; a key change requires a new device entry. Within a batch, the
first record of a device reads and updates the slot and later records of the same device find it warm, so
the rule costs one storage update per device and batch (Section 7). It requires the aggregator to submit
each device’s records in timestamp order: a record withheld until a later one of the same device has been
anchored can no longer be anchored, a liveness limitation that a window of recently accepted timestamps
would remove at a slightly higher cost.
5 Authentication designs
5.1 ECDSA over secp256k1 (K1)
The device signsh=keccak256(d∥M), where the domaindprevents replay across chains and deploy-
ments. The contract rejects high-ssignatures (malleability), recovers the signer withecrecover(3,000
gas), and checks and advances the signer’s timestamp, whose slot doubles as the registry entry. A signa-
ture is 65 bytes.
5.2 ECDSA over P-256 (R1)
The device signsh=SHA-256(d∥M), as supported by secure elements. The contract checks the public
key(q x,q y)against the registry commitment and calls the EIP-7951 precompile (6,900 gas). In a batch,
public keys are supplied once per device. A signature is 64 bytes.
5.3 BLS aggregate signatures (BLS)
Each devicejholdssk j ∈Z ∗r andpk j =sk jP2 ∈G 2, and signsσ i =sk j H(d∥c j∥Mi), whereHis the
RFC 9380 hash-to-curve suiteBLS12381G1 XMD:SHA-256 SSWU RO [32] andc j the device’s registry
commitment. The aggregator sums all signatures of a batch intoσ= ∑i σi (48 bytes compressed), so that
calldata contains one signature per batch. The contract hashes every message toG 1, sums the hashes per
device and checks
e(σ,−P 2)· ∏
j
e

∑
i∈Ij
H(c j∥Mi),pk j

=1,(1)
a pairing check withm+1 pairs formdevices. Aggregation is secure for distinct messages [30], which
timestamps and checksums guarantee, and rogue-key attacks are prevented because keys are registered
by the KGC. Our Solidity hash-to-curve implementsexpand message xmdwith the SHA-256 precom-
pile, reduces the two 64-byte outputs modulopwith MODEXP and maps them with the EIP-2537
6

Algorithm 1:MLA V-IBS
1Setup(KGC):
2s $← −Z∗r ;returnP pub,1 =sP 1,P pub,2 =sP 2;
3Extract(RID)(KGC):
4t $← −Z∗r ;SID 1 ←tP 1;SID 2 ←RID⊕H 0(sSID 1);ID←SID 1∥SID2;
5Q ID ←H 1(ID);returnsk ID =sQ ID ;// onlysk ID is stored on the device
6Sign(sk ID,M):
7ρ $← −Z∗r ;U←ρQ ID;h←H 2(ID,M,U);return(U,V= (ρ+h)sk ID);
8Verify(ID,M,U,V):
9returne(V,P 2)·e(U+H 2(ID,M,U)Q ID,−P pub,2) ?=1;
10Trace(ID)(KGC):
11returnRID=SID 2 ⊕H 0(sSID 1);
MAP FP TO G1precompile, which includes cofactor clearing; we verified that its output equals the refer-
ence implementation [45].
5.4 Pseudonymous identity-based signature (IBS)
Why not the original MLA V signature.The MLA V design [16], specified in [17], used a signature
of the IBV family [18]: a device holdsPrK 1 =c 1 SID1 andPrK 2 =c 2 H1(SID1∥SID2), derived inside
the device from the KGC’s master keysc 1,c 2, and signsσ=vPrK 1 +wPrK 2 with public scalarsv,w
computed from the record; batches are verified by summing signatures. The signature is deterministic
and linear in the two keys, so any two signatures of a device form a 2×2 linear system whose solution
is(PrK 1,PrK 2), enabling universal forgery for that device; plain-sum batching accepts pairs of invalid
signatures whose errors cancel; and every device stores the master keys. Weaknesses of this kind have
been reported for the IBV family [19]; our released code reproduces both attacks against the scheme as
specified in [17]. The scheme of [17] therefore does not provide unforgeability and should not be used;
we replace it.
MLA V-IBS.We use a pseudonymous variant of the Cha–Cheon IBS [35] over the Type-3 pairing
e:G 1 ×G 2 →G T of BLS12-381 (Algorithm 1). The KGC holdssand publishesP pub,1 =sP 1 and
Ppub,2 =sP 2. A pseudonymID=SID 1∥SID2 withSID 1 =tP 1 andSID 2 =RID⊕H 0(sSID 1)hides the
real identityRIDfrom everyone but the KGC; the device stores onlysk ID =sH 1(ID). A signature
is(U,V) = (ρQ ID,(ρ+h)sk ID)with freshρandh=H 2(ID,M,U), whereMincludes the domaind
as in all designs, and is valid ife(V,P 2)e(U+hQ ID,−Ppub,2) =1. Unforgeability follows from the
Cha–Cheon proof [35] adapted to Type-3 pairings, under the assumption that computingabP 1 from
(P1,aP1,bP1,P2,aP2)is hard; publishingP pub,1 lets the reduction simulate extraction queries (Section A).
Batch verification.With randomℓ-bit weightsδ i, the small-exponent test [37] (B2) checks
e

∑
i
δiVi,P 2

·e

∑
i
δiUi +∑
i
δihi QIDi,−P pub,2

=1.(2)
7

BecauseQ ID depends only on the pseudonym, the identity terms can be collected per pseudonym,
∑i δihiQIDi = ∑ j

∑i∈Ij δihi

Q j, which we callpseudonym-groupedbatching (B3). It shrinks the sec-
ond MSM from 2nton+mpoints and the number of evaluations ofH 1 fromntom. B3 is an algebraic
regrouping of the small-exponent test, not a new verification technique, and its saving rests on many
records sharing a pseudonym. Records signed under one pseudonym are linkable to each other, though
not to the device’s real identity. A device that changes its pseudonym for every record to make its records
unlinkable gains nothing from grouping (m=n) and needs a new on-chain registration per pseudonym
(Section 7). Grouping and unlinkability between records are therefore alternatives. In the supply chain
finance setting, the records of a farm are meant to be linked to its loan, so pseudonyms are long-lived and
grouping applies; where records must be unlinkable, B2 is the relevant baseline.
Theorem 1(Batch soundness).Let all U i,Vi ∈G 1 and let theδ i be independent and uniform over a set
∆of integers that are distinct modulo r. If some signature in the batch is invalid, B2 and B3 accept with
probability at most1/|∆|.
Proof.Letγ i =e(V i,P2)e(Ui +h iQIDi,−Ppub,2), so that signatureiis valid iffγ i =1. By bilinearity both
checks compute ∏i γ δii . Writingγ i =g αi for a generatorgofG T , acceptance means ∑i αiδi ≡0(modr).
Ifα j ̸=0, then for any fixed{δ i}i̸=j exactly one residue ofδ j satisfies the congruence, whichδ j hits with
probability at most 1/|∆|.
Off-chain, we use 64-bit odd weights (error 2 −63). A contract has no private randomness, so we de-
rive 128-bit weights asδ i =⌊keccak256(keccak256(calldata),i)/2 128⌋in the Fiat–Shamir style [46]; an
adversary who triesqbatches succeeds with probability at mostq/2 128. Subgroup membership, required
by Theorem 1, is checked by our deserializer and by the EIP-2537 MSM and pairing precompiles. On-
chain, the KGC registersc ID =keccak256(Q ID), submitters provideQ ID, andH 2 is keccak256 modulor
over the domaind,c ID,MandU.
6 Methodology
On-chain.Contracts are written in Solidity 0.8.28 (optimizer, 200 runs, via-IR) and executed as signed
EIP-1559 transactions on the EthereumJS VM 10.1.3 [47] configured with the Osaka hardfork, which
enforces EIP-2537, EIP-7623, EIP-7825, EIP-7883 and EIP-7951. We report the total gas of each trans-
action, including the 21,000 intrinsic gas and calldata, in a block with a non-zero block number so that
every anchoring write is a fresh zero-to-non-zero storage write. Signatures and test data are produced
with independent JavaScript implementations [45]. Individual submission is measured as the mean of
five transactions. Batches containnrecords fromm=min(n,10)devices, round-robin. The largest batch
under the EIP-7825 cap is found by bisection with actual transactions. Gas figures are deterministic.
Off-chain.Device signing and aggregator verification are measured in Python 3.12 on a single virtual
core of an Intel Xeon processor at 2.10 GHz (Linux, glibc 2.39), the reference machine: ECDSA with
libsecp256k1[48] (viacoincurve21.0) and OpenSSL 3.5 (viacryptography46.0), and BLS and
MLA V-IBS with the Rustarkworkslibrary [49] (py arkworks bls123810.5.0). We report medians of
7 repetitions (140 for signing). To check that the results do not depend on the processor, we repeated all
off-chain measurements on GitHub-hosted runners with an ARM64 (Arm Neoverse-N2) and an x86-64
8

Table 1: Gas per record under Osaka rules (m=min(n,10)devices). “Individual” is one transaction per
record that verifies its signature and stores the record hash; batches verify all signatures and store one
Merkle root. The last column is the largest batch that fits under the EIP-7825 cap of 224 gas.
Design Individualn=10n=100n=200n=500 Max.n
T0 root only (no authentication) – 4,548 455 – 91 –
T1 record hashes (no authentication) – 5,762 1,644 – 1,319 –
K1 ECDSA secp256k155,47517,896 9,492 9,0788,963 1,754
R1 ECDSA P-256 63,788 27,066 15,129 14,580 14,542 1,105
BLS aggregate 187,118 82,455 34,927 35,230 – 411
IBS small-exponent (B2) 176,943 58,480 31,242 30,296 31,202 533
IBS pseudonym-grouped (B3) 176,943 58,945 25,776 24,228 24,634 662
(AMD EPYC 9V74) processor (Table 3): signing times stay within 18% and per-record verification times
within 31% of the reference machine, and the ranking of the designs is unchanged. These are server-class
cores, so we also measured two Raspberry Pi boards, recording each board’s clock and throttling state
before and after every run: every timing in Table 3 is 4.4 to 7.3 times slower on a Raspberry Pi 4 than on
the reference machine and 11 to 16 times slower on a Raspberry Pi Zero 2 W, the class of board used in
the IoP sensors, and the ranking of the designs is again unchanged. To measure aggregator throughput
on several cores, worker processes verify independent, pre-signed batches of 1,000 records concurrently,
and throughput is the total number of records divided by the slowest worker’s time. On the Raspberry
Pi boards every measurement was repeated 10 times, with a pause between runs, and we report medians
across runs together with the largest coefficient of variation.
Correctness checks.For every design, valid batches are accepted and batches with a tampered record
are rejected; the on-chain hash-to-curve matches RFC 9380; and the MLA V-IBS verifiers reject the
cancellation batch that the plain-sum verifier accepts, both off-chain and on-chain.
7 Results
7.1 Anchoring without authentication (T0, T1)
Posting only a Merkle root (T0) costs 45,507 gas per transaction regardless ofn, i.e. 91 gas per record
forn=500. Posting all record hashes and computing the root on-chain (T1) costs 659,610 gas for 500
records (1,319 gas per record), and 1,298,080 gas for 1,000 records. Storing each record hash in its own
storage slot instead costs 11,390,122 gas for 500 records (22,780 per record), 17 times more than T1, and
1,000 per-slot writes exceed the transaction cap. Merkle anchoring is therefore the single largest saving
available, and it is independent of the signature scheme.
7.2 On-chain authentication (T2)
Table 1 and Figure 3 summarize the measurements.
Individual submission.Submitting each record in its own transaction costs between 55,475 (K1) and
187,118 gas (BLS). For ECDSA the cost is dominated by the 21,000 intrinsic gas, the 22,100-gas storage
write of the record hash and the update of the device’s timestamp, while signature verification itself costs
9

1 10 100 1,000
Records per transaction, n
1,000
10,000
100,000
1,000,000Gas per record
(a)
BLS aggregate
IBS small-exponent (B2)
IBS pseudonym-grouped (B3)
R1 ECDSA P-256
K1 ECDSA secp256k1
T1 anchoring, no authentication
0 500 1,000 1,500 2,000
Largest batch per transaction (records)
K1
R1
IBS-B3
IBS-B2
BLS
1,754
1,105
662
533
411
(b)
EIP-7825 cap: 16,777,216 gas
Figure 3: (a) Gas per record versus records per transaction (log–log,m=min(n,10)devices); curves end
where the next measured batch size exceeds the per-transaction cap. (b) Largest batch per transaction
under the EIP-7825 cap, found by bisection with actual transactions.
only 3,000 (K1) or 6,900 gas (R1). For the pairing-based designs, the pairing check (102,900 gas for two
pairs) dominates.
Batching.Batching amortizes the intrinsic cost, the storage write and, for pairing-based designs, the
pairing check. Per-record cost falls steeply up to about 100 records and is flat beyond (Figure 3a); for very
large batches it rises slightly because EVM memory expansion is priced quadratically. Atn=500, K1
costs 8,963 gas per record, 6.2 times less than individual submission; R1 costs 14,542 gas (1.6 times K1).
MLA V-IBS with pseudonym grouping costs 24,634 gas (2.7 times K1), 21% less than the ungrouped
small-exponent verifier. BLS aggregation costs 35,230 gas atn=200 (3.9 times K1), although it puts
only one signature per batch in calldata.
Where the gas goes.For K1, a batched record consists of about 224 bytes of calldata, oneecrecover
(3,000 gas), two keccak256 hashes and a warm read and update of the device’s timestamp. For BLS,
hashing a message toG 1 costs 30,869 gas in our implementation. Its precompile floor is about 13,000
gas (twoMAP FP TO G1calls at 5,500 gas, two MODEXP calls at the post-Fusaka minimum of 500
gas, one G1 addition and five SHA-256 calls), the rest being memory and encoding overhead. Even an
implementation reaching this floor would cost more than 15,000 gas per record and remain above K1.
For MLA V-IBS, each record contributes two points to G1 MSMs in B3 (three in B2), about 12,500 gas
at the maximum MSM discount, plus about 420 bytes of calldata for two uncompressed EIP-2537 points
and the record.
Fixed costs.Deployment costs 483,408 (K1), 683,081 (R1), 942,740 (BLS) and 1,195,856 gas (IBS);
registering a device costs 44,097–69,751 gas, including the initialisation of its timestamp slot.
7.3 The transaction cap and block capacity
Since Fusaka, a transaction cannot exceed 2 24 gas. The largest batch is 1,754 records for K1, 1,105
for R1, 662 for IBS-B3, 533 for IBS-B2 and 411 for BLS (Figure 3b). Larger batches must be split;
each additional transaction adds the fixed batch overhead of about 45,000–200,000 gas depending on
10

Table 2: Off-chain costs (single core). Verification is per record for a batch ofnrecords from 10 devices;
IBS uses pseudonym-grouped batching (B3), withH 1 evaluated once per pseudonym. Signature sizes
are compressed encodings.
Sign (ms) Signature Verify, ms per record
Design bytesn=1n=100n=1000
K1 ECDSA secp256k1 0.024 65 0.038 0.040 0.036
R1 ECDSA P-256 0.026 64 0.080 0.087 0.072
BLS aggregate 0.400 48 per batch 1.771 0.316 0.256
IBS grouped (B3) 0.398 96 2.221 0.093 0.038
Table 3: Off-chain timings on five machines: the reference machine of Table 2 (Intel Xeon at 2.10 GHz,
one vCPU), GitHub-hosted runners with an ARM64 Arm Neoverse-N2 and an x86-64 AMD EPYC
9V74 (four vCPUs each), a Raspberry Pi 4 Model B (four Arm Cortex-A72 cores at 1.8 GHz, 2 GB
RAM) and a Raspberry Pi Zero 2 W (four Arm Cortex-A53 cores at 1 GHz, 512 MB RAM), the class of
board used in the IoP sensors. Single-threaded medians with the same package versions; the Raspberry
Pi columns are medians of 10 runs each, whose coefficients of variation are at most 2.3%; neither board
throttled during any run.
Xeon Neoverse-N2 EPYC 9V74 Pi 4 Pi Zero 2 W
(reference) (ARM64) (x86-64) (Cortex-A72) (Cortex-A53)
Device signing (ms per record)
ECDSA secp256k1 (K1) 0.024 0.028 0.026 0.128 0.287
ECDSA P-256 (R1) 0.026 0.025 0.024 0.128 0.344
BLS 0.400 0.433 0.399 2.384 5.699
MLA V-IBS 0.398 0.434 0.401 2.293 5.497
Aggregator verification (ms per record, n=1,000)
ECDSA secp256k1 (K1) 0.035 0.045 0.047 0.258 0.582
ECDSA P-256 (R1) 0.072 0.067 0.064 0.321 0.799
BLS 0.256 0.258 0.259 1.306 3.028
MLA V-IBS 0.038 0.035 0.036 0.170 0.403
MLA V-IBS,Qcached 0.031 0.033 0.034 0.157 0.373
the design, which is small relative to the batch. With the default block gas limit of 60 million [14], and
ignoring other traffic, one block could anchor about 6,700 K1-authenticated records, 4,100 R1, 2,500
IBS-B3, 1,700 BLS or 45,000 records at trust level T1.
7.4 Device and aggregator costs
Table 2 reports off-chain costs. Pairing-based signing is about 16 times slower than ECDSA signing on
the reference machine (0.40 versus 0.024 ms), about 19 times slower on a Raspberry Pi 4 and about 20
times slower on the Raspberry Pi Zero 2 W of the IoP sensor (5.7 versus 0.29 ms, Table 3). The relative
gap is large, but in absolute terms a few milliseconds per record are negligible for a sensor that signs
one record every few minutes, so for battery-powered devices how often they sign and transmit matters
more than which of these schemes they use. For the aggregator, ECDSA verification costs 0.036 ms (K1)
and 0.072 ms (R1) per record at any batch size. Batch verification lets MLA V-IBS match K1 for large
batches (0.038 ms per record atn=1000), whereas BLS remains 7 times slower than K1 because every
message must be hashed to the curve. Pseudonym grouping matters most whenH1 must be evaluated: for
1,000 records from 50 pseudonyms, B3 takes 64 ms against 351 ms for B2; withH 1 cached, the saving
11

Table 4: Aggregator throughput: records verified per second with 1, 2 and 4 worker processes, each
verifying pre-signed batches of 1,000 records from 10 devices, and the speedup over one worker. The
Xeon runner has two physical cores with two hardware threads each; the other machines have four cores
with one thread each. Raspberry Pi rows are medians of 10 runs, whose coefficients of variation are at
most 1.1%; neither board throttled.
Machine, workers (w.) K1 R1 BLS IBS (Qcached)
Neoverse-N2 (4 cores), 1 w. 22,047 14,889 3,874 30,920
4 w. 87,464 (3.97×) 59,098 (3.97×) 15,391 (3.97×) 123,347 (3.99×)
Xeon 6975P-C (2 cores×2 threads), 1 w. 28,884 15,543 4,496 34,627
2 w. 56,927 (1.97×) 29,717 (1.91×) 8,837 (1.97×) 69,271 (2.00×)
4 w. 68,025 (2.36×) 35,435 (2.28×) 11,108 (2.47×) 87,604 (2.53×)
Raspberry Pi 4 (4 cores), 1 w. 3,884 3,123 766 6,376
4 w. 15,158 (3.90×) 12,075 (3.86×) 3,023 (3.95×) 24,439 (3.83×)
Raspberry Pi Zero 2 W (4 cores), 1 w. 1,718 1,250 330 2,683
4 w. 6,812 (3.96×) 4,421 (3.54×) 1,320 (4.00×) 10,471 (3.90×)
1 10 100 1,000
Distinct pseudonyms in the batch, m
0
10
20
30
40
50
60
70Verification time (ms)
n = 1,000 records, QID cached
IBS small-exponent (B2)
IBS pseudonym-grouped (B3)
Figure 4: Off-chain MLA V-IBS batch verification time for 1,000 records versus the number of distinct
pseudonyms (QID cached). Grouping helps most when few pseudonyms produce many records.
is 35–40% for few pseudonyms and vanishes when every record comes from a different pseudonym
(Figure 4).
Aggregator throughput on several cores.An aggregator verifies independent batches in parallel, so
its throughput scales with the number of physical cores (Table 4): with four worker processes on four
cores, every design verifies 3.8 to 4.0 times as many records per second as with one, on the Neoverse-
N2 server cores as on both Raspberry Pi boards; only R1 on the Pi Zero 2 W scales less (3.54 times,
consistently across 10 runs), whereas on the Xeon runner the third and fourth workers share two cores
with the first two and hardware threads add only 19–26%. Even one core is far from being the bottleneck.
At about 6,700 K1-authenticated records per 12-second block (Section 7.3), the chain accepts about 560
records per second, whereas one Neoverse-N2 core verifies about 22,000; for BLS, about 140 records per
second against about 3,900, and for IBS-B3 about 210 against about 31,000. Even a Raspberry Pi Zero
2 W with four workers verifies 6,812 K1, 1,320 BLS and 10,471 IBS records per second, 9 to 50 times
what the chain can accept. The cost of authenticating IoT records is therefore set by the chain’s gas, not
by the aggregator’s computation, and even a low-cost board can serve the full capacity of the chain.
12

Table 5: The same transactions under the rules before Pectra (Cancun), after Pectra (Prague) and after
Fusaka (Osaka): gas per record,m=min(n,10)devices. “n/a”: the precompile the design needs is not
available.
Design Cancun Prague Osaka
T1 record hashes,n=500 1,286 1,320 1,319
K1 ECDSA secp256k1,n=500 8,964 8,963 8,963
R1 ECDSA P-256,n=500 n/a n/a 14,542
BLS aggregate,n=100 n/a 34,328 34,927
BLS aggregate,n=500 n/a 42,251 above cap
IBS pseudonym-grouped (B3),n=500 n/a 24,633 24,634
Largest transaction 36 M (block) 36 M (block) 2 24 (EIP-7825)
7.5 Before and after Pectra and Fusaka
To isolate what the two upgrades changed, we executed the same contracts and transactions under the
Cancun rules that preceded Pectra and the Prague rules of Pectra (Table 5). Their main effect is on
feasibility rather than cost. Before Pectra, only K1 can be verified with a precompile; Pectra’s BLS12-381
precompiles (EIP-2537) make BLS and MLA V-IBS possible, and Fusaka’s P-256 precompile (EIP-7951)
adds R1. Designs available under several rules cost almost the same under each: K1 within 0.01%, and
IBS within a few gas. Two repricings are visible. Fusaka’s MODEXP repricing (EIP-7883) adds 600 gas
to every hash toG 1 (30,269 against 30,869 gas), about 1.7% per BLS record atn=100. Pectra’s calldata
floor (EIP-7623) raises T1, which posts data with little execution, by 2.6%, and binds for none of the
authenticated designs. The largest change is the per-transaction cap of Fusaka. Before it, a transaction
was bounded only by the block gas limit, 36 million gas when Pectra activated, so a BLS batch of 500
records (42,251 gas per record) and 1,000 per-slot writes (22.8 million gas) were possible; under Osaka
both exceed 224 gas. The cap therefore matters for designs whose cost grows with the batch, not for the
cheapest ones.
7.6 Alternatives we did not implement
Record hashes in a blob (T1).A blob carries 4,096 field elements of which 31 bytes each are usable,
enough for 3,968 record hashes [50]. Its gas is priced in a separate market whose base fee is, since
Fusaka, bounded below so that a blob costs at least as much as 8,192 gas of execution [51]. Posting the
hashes of a T1 batch in a blob, and only the root and the blob’s versioned hash in the transaction, would
cost about one T0 transaction with a second storage write plus one blob, about 76,000 gas for 3,968
records, i.e. roughly 20 gas per record at the reserve price against 1,319 for T1 (an estimate; we did not
execute blob transactions). The saving has two costs. A contract can read only a blob’s commitment,
not its contents, so it can no longer compute the root itself and must accept the aggregator’s claim that
the root matches the blob, unless each record is opened with the point-evaluation precompile. And blob
contents are kept for only about 18 days, after which the publicity that defines T1 depends on archivers.
PeerDAS [52], introduced with Fusaka, lets the network carry more blobs, which makes this option more
available but not more trustworthy. A blob is therefore a cheaper data-availability layer for T0 rather
than a substitute for T1.
13

Table 6: Properties of the designs. The KGC registers each device once; in K1, R1 and BLS the device
generates its own key, whereas in IBS the KGC derives it.
Property K1 R1 BLS IBS
Supported by device secure elements partial yes no no
Key material per device on-chain address commitment commitment commitment
Public-key distribution needed off-chain yes yes yes no
Signatures in calldata per record per record one per batch per record
Pseudonymity with KGC tracing via KGC records via KGC records via KGC records built in
Device compromise exposes own key own key own key own key
KGC can sign as a registered device no no no yes (escrow)
Proof-verified aggregation (T2).Instead of verifying every signature on-chain, an aggregator can
verify the signatures and the replay rule off-chain and prove in a succinct argument that the posted root
commits only to records signed by registered devices with fresh timestamps. A Groth16 proof [53] over
BN254 is verified with one four-pair pairing check, 181,000 gas under EIP-1108 [54], plus a 6,000-gas
scalar multiplication per public input. With the root and the old and new commitments to the device
timestamps as public inputs, the 21,000-gas base, a 256-byte proof and two storage writes, a batch would
cost about 260,000 gas regardless of its size (an estimate): less than K1 from about 30 records per batch,
and about 260 gas per record atn=1,000. The per-device timestamps must move into the committed
state that the proof updates; otherwise the per-device storage updates of Section 4 remain. The costs
move off-chain. secp256k1 signatures must be verified with non-native arithmetic inside the circuit,
which makes proving far more expensive than verifying the signatures directly and adds the proving time
to the anchoring latency; Groth16 also needs a trusted setup per circuit, and pairing-based proofs are not
post-quantum. Proof-verified aggregation is attractive for large batches when an aggregator with proving
hardware is available; implementing it and measuring the prover are left to future work.
7.7 Qualitative properties
Table 6 compares properties that do not show up in gas. All designs can provide pseudonymity towards
outsiders with tracing by the KGC, ECDSA and BLS by having the KGC issue and record per-device
keys; the identity-based design makes this mapping cryptographic and removes the need to distribute
public keys off-chain, an advantage that the on-chain registry largely neutralizes.
Key escrow.The identity-based design pays for this convenience with key escrow. The KGC computes
every device key,sk ID =sH 1(ID), and can therefore sign records in the name of any device, and such
signatures are indistinguishable from the device’s own. Since the consortium that operates the KGC
includes the lender, the lender could fabricate the production records that release its own payments,
which is exactly what on-chain authentication (T2) is meant to rule out. In the other designs the device
generates its key and the KGC only registers it; because the registry accepts each device once, the KGC
can add devices but cannot sign as an existing one or replace its key, and every registration is a public
transaction. The escrow can be removed at additional cost by splitting the master key among consortium
members with threshold key extraction [34] or by certificateless signatures, in which part of the key
is chosen by the device [55]; we did not implement either. Built-in pseudonymity with conditional
traceability is therefore the only advantage of the identity-based design, and it comes at the price of
14

trusting the KGC with the authenticity of every record.
8 Findings and guidelines
Finding1.Anchoring is the first lever. Merkle anchoring reduces the cost of recording a hash from
22,780 gas (one storage slot per record) to 1,319 gas when all hashes are published (T1), and to a
constant 45,507 gas per batch when only the root is posted (T0). Any design that writes one storage slot
per record should be avoided.
Finding2.On-chain authentication costs a factor of six to seven. Moving from T1 to T2 with the cheapest
scheme raises the cost from 1,319 to 8,963 gas per record. Whether this is worthwhile depends on
whether the aggregator can be trusted; in a consortium in which the lender itself operates the aggregator,
T1 may be sufficient, whereas T2 is needed when records serve as evidence towards parties outside the
consortium.
Finding3.For on-chain authentication, ECDSA over secp256k1 with batching is the cheapest design
measured, at 8,963 gas per record and up to 1,754 records per transaction. P-256 costs 1.6 times as much
but matches the hardware that IoT devices typically embed. Pairing-based designs cost 2.7–3.9 times as
much on-chain and are 16 times slower to sign with.
Finding4.BLS aggregation does not pay off on-chain for IoT records. Aggregation removes per-record
signatures from calldata, but every message must still be hashed to the curve on-chain, which costs more
than verifying an ECDSA signature. It also yields the smallest batches under the transaction cap (411
records).
Finding5.Pairing-based identity-based signatures are justified by properties, not by cost. Their batch
verification makes them competitive off-chain with ECDSA for large batches, and pseudonym grouping
reduces their on-chain cost by 22% and their off-chain cost by up to 5.5 times when identities must be
hashed. They are worth their premium when built-in pseudonymity with conditional traceability is a
requirement, and only if the scheme is chosen with care: signatures of the IBV family, as used in the
original MLA V prototype, are insecure.
Converting gas to fees.The fee of a record is its gas multiplied by the prevailing gas price; for example,
at a base fee ofggwei and an ether price ofEUSD, a K1 record in a batch costs 8,963·g·E·10 −9 USD.
Because the ranking of designs depends only on gas, our conclusions hold for any fee level. On rollups
that execute the same precompiles, execution gas is the same, while the cost of calldata depends on the
rollup’s data-availability pricing; designs with less calldata per record, such as BLS aggregation, would
gain relative to ECDSA there.
9 Limitations
Our implementations are straightforward Solidity without hand-written assembly beyond buffer han-
dling; optimized code, in particular for BLS hash-to-curve, would lower the absolute costs of the pairing-
based designs, but our precompile-floor analysis indicates that it would not change the ranking. Off-chain
timings were taken on server cores and on two Raspberry Pi boards of the class used in the IoP sensors,
not on sensors deployed in the field, and we did not measure energy. The blob-based and proof-verified
15

alternatives of Section 7.6 are estimated from protocol prices, not implemented. The workload is syn-
thetic, with 10 devices per batch; we have not evaluated localization of invalid signatures in a failed batch,
which requires bisection [39], nor deployment on rollups. Finally, gas schedules change with network
upgrades; our harness makes it straightforward to repeat the study under future rules, as we did for the
rules that preceded Pectra and Fusaka (Section 7.5).
10 Conclusion
We measured what it costs to authenticate and anchor IoT records on Ethereum after the Pectra and
Fusaka upgrades. Merkle anchoring and batching are the decisive levers, reducing the cost per record by
more than an order of magnitude. Among authenticated designs, batched secp256k1 ECDSA is the cheap-
est at 8,963 gas per record, while the new BLS12-381 precompiles make pairing-based designs feasible
but not competitive in cost: they are justified only by additional properties such as built-in pseudonymity
with conditional traceability, which the identity-based design obtains at the price of key escrow at the
KGC. For agricultural supply chain finance, this suggests anchoring record hashes at trust level T1 within
a lender-operated consortium and adding batched ECDSA or P-256 authentication when records must
convince outside parties. Future work includes energy measurements on deployed sensors, an imple-
mentation of proof-verified aggregation with a measurement of its prover (Section 7.6), deployment on
rollups and an assessment of privacy-preserving credit evidence based on zero-knowledge proofs.
Acknowledgements
The MLA V architecture, the camera prototype and the supply chain finance workflow on which this study
builds were developed with Jingze Wu, Seth Austin Harding, Chien-Lung Lin, San-Tai Wang and Bo-
Yen Tang at National Taiwan University [16, 17]; the authors gratefully acknowledge their contributions.
The camera prototype and the first smart-contract mock-up of the platform originated in the final project
“Agricultural Blockchain: A Reconceptualized Approach to Organic” of the 2019 blockchain course at
National Taiwan University; we thank its student team.
Declarations
Funding.This research did not receive any specific grant from funding agencies in the public, com-
mercial, or not-for-profit sectors.
Competing interests.Ming-Fong Sie was a co-founder and the chief executive officer, and Shih-wei
Liao a co-founder, of Conzen.AI, a company that developed privacy-preserving computation for financial
compliance and supply chain traceability and was dissolved in October 2026. The company did not fund
this study and had no role in its design, analysis or reporting. The other authors declare no competing
interests.
Ethics approval and consent to participate.Not applicable.
Consent for publication.Not applicable.
16

Data and code availability.The contracts, benchmark harnesses, scripts reproducing the attacks
on the original MLA V signature, and all raw measurements are available athttps://github.com/
Siemingfong/iot-auth-cost-study(release v1.1), archived at Zenodo under DOI 10.5281/zen-
odo.24328819.
Author contributions. Ming-Fong Sie: conceptualization, methodology, software, investigation, for-
mal analysis, data curation, visualization, writing – original draft.Jian-Hong Chen: investigation, soft-
ware, validation, writing – review & editing.Gwan-yen Lin: investigation, validation, writing – review
& editing.Yen-Shuo Chen: formal analysis, validation, writing – review & editing.Shih-wei Liao:
conceptualization, supervision, resources, writing – review & editing.
Use of generative AI.The authors used Claude (Anthropic) to assist with writing the benchmark code
and Solidity contracts, running the experiments, analysing data, preparing figures, checking the security
argument of the appendix, and drafting and editing the text. All code, measurements and text were
reviewed by the authors, who take full responsibility for the content of the publication.
A Unforgeability of MLA V-IBS (proof sketch)
Model and assumption.We use existential unforgeability under adaptive chosen-message and chosen-
identity attacks (EUF-ID-CMA): the adversary receives the public parameters, may adaptively queryH1,
H2, a key-extraction oracle and a signing oracle, and wins by outputting a valid signature on(ID ∗,M ∗)
for anID ∗ it never extracted and a pair(ID ∗,M ∗)it never submitted to the signing oracle. The reduc-
tion targets the following computational co-Diffie–Hellman problem in the Type-3 setting [56]: given
(P1,aP1,bP1,P2,aP2)for uniforma,b∈Z ∗r , computeabP 1. The elementaP 2, which the plain computa-
tional Diffie–Hellman problem inG 1 lacks, is needed to publishP pub,2.
Reduction.The argument follows Cha and Cheon [35] in the random oracle model forH 1 andH 2.H 1
is instantiated with the random-oracle variant of the RFC 9380 hash-to-curve, which is designed to be
indifferentiable from a random oracle [32]. The simulator setsP pub,1 =aP 1 andP pub,2 =aP 2, guesses
the target pseudonymID ∗ among theq H1 queries toH 1, programsH 1(ID∗) =bP1, and answers otherH 1
queries withx IDP1 for uniformx ID. Extraction queries forID̸=ID ∗ are answered withsk ID =x ID Ppub,1;
an extraction query forID ∗ means the guess was wrong, and the simulator aborts. Signing queries for
any pseudonym are answered without the key: the simulator sampleshwith the distribution ofH 2’s
output (for keccak256 reduced modulor, which is not exactly uniform onZ ∗r , by reducing a uniform
256-bit string), samplesy $← −Z∗r , setsU=yP 1 −hQ ID andV=yP pub,1, and programsH 2(ID,M,U) =h;
indeede(V,P 2) =e(yP1,aP2) =e(U+hQ ID,Ppub,2). SinceUis uniform inG 1, the programmed point is
already defined only with probability at most(q H2 +q S)q S/r, in which case the simulator aborts. From a
forgery(U,h,V)onID ∗, the forking lemma [57] yields a second forgery with the sameUand a different
h′. Every valid signature satisfiese(V,P 2) =e(U+hQ ID∗,aP2), and since the pairing is non-degenerate
this meansV=a(U+hQ ID∗), whatever way the forger produced it; henceV−V ′ = (h−h ′)aQ ID∗ and
(h−h ′)−1(V−V ′) =abP1. SinceH 2’s outputs lie inZ ∗r and collide with negligible probability,h−h ′ is
invertible modulor.
17

Tightness and batches.The reduction loses a factor of aboutq H1 for guessingID ∗ and, through the
forking lemma, roughly squares the adversary’s advantage, so it is not tight, as in Cha and Cheon’s
analysis. Combined with Theorem 1, an accepted batch contains a forged record only if the adversary
breaks the unforgeability of a single signature or the batch test, so accepted batches are unforgeable up
to the sum of both bounds, the bound of Theorem 1 counted once per submitted batch; on-chain,H 2
additionally binds the domaind, which prevents reuse of signatures across chains and contracts.
References
[1] Mischa Tripoli and Josef Schmidhuber. Emerging opportunities for the application of blockchain
in the agri-food industry.FAO and ICTSD: Rome and Geneva, 2018.
[2] Andreas Kamilaris, Agusti Fonts, and Francesc X. Prenafeta-Bold ´u. The rise of blockchain tech-
nology in agriculture and food supply chains.Trends in Food Science & Technology, 91:640–652,
2019.
[3] Yingli Wang, Jeong Hugh Han, and Paul Beynon-Davies. Understanding blockchain technology
for future supply chains: a systematic literature review and research agenda.Supply Chain Man-
agement: An International Journal, 24(1):62–84, 2019.
[4] Miguel Pincheira Caro, Muhammad Salek Ali, Massimo Vecchio, and Raffaele Giaffreda.
Blockchain-based traceability in agri-food supply chain management: A practical implementation.
In2018 IoT Vertical and Topical Summit on Agriculture – Tuscany (IOT Tuscany), pages 1–4, Pis-
cataway, NJ, USA, 2018. IEEE.
[5] Shabir Ahmad Bhat, Nen-Fu Huang, Ishfaq Bashir Sofi, and Muhammad Sultan. Agriculture-
food supply chain management based on blockchain and IoT: A narrative on enterprise blockchain
interoperability.Agriculture, 12(1):40, 2022.
[6] Jiri Chod, Nikolaos Trichakis, Gerry Tsoukalas, Henry Aspegren, and Mark Weber. On the financ-
ing benefits of supply chain transparency and blockchain adoption.Management Science, 66(10):
4378–4396, 2020.
[7] EIP-7600: Hardfork meta – pectra. Ethereum Improvement Proposals,https://eips.ethereum.
org/EIPS/eip-7600, 2025.
[8] EIP-2537: Precompile for BLS12-381 curve operations. Ethereum Improvement Proposals,
https://eips.ethereum.org/EIPS/eip-2537, 2020. Activated on mainnet in the Pectra up-
grade (2025).
[9] EIP-7623: Increase calldata cost. Ethereum Improvement Proposals,https://eips.ethereum.
org/EIPS/eip-7623, 2025.
[10] EIP-7607: Hardfork meta – fusaka. Ethereum Improvement Proposals,https://eips.ethereum.
org/EIPS/eip-7607, 2025.
[11] Ethereum Foundation Protocol Coordination Team. Fusaka mainnet announcement.https://
blog.ethereum.org/2025/11/06/fusaka-mainnet-announcement, 2025.
18

[12] EIP-7951: Precompile for secp256r1 curve support. Ethereum Improvement Proposals,https:
//eips.ethereum.org/EIPS/eip-7951, 2025.
[13] EIP-7883: Modexp gas cost increase. Ethereum Improvement Proposals,https://eips.
ethereum.org/EIPS/eip-7883, 2025.
[14] EIP-7935: Set default gas limit to 60m. Ethereum Improvement Proposals,https://eips.
ethereum.org/EIPS/eip-7935, 2025.
[15] EIP-7825: Transaction gas limit cap. Ethereum Improvement Proposals,https://eips.
ethereum.org/EIPS/eip-7825, 2025.
[16] Jingze Wu, Ming-Fong Sie, Seth Austin Harding, Chien-Lung Lin, San-Tai Wang, and Shih-wei
Liao. Multi-layer aggregate verification for IoT blockchain. In2021 3rd Conference on Blockchain
Research & Applications for Innovative Networks and Services (BRAINS), pages 43–44, Piscataway,
NJ, USA, 2021. IEEE. doi: 10.1109/BRAINS52497.2021.9569817.
[17] Ming-Fong Sie, Jingze Wu, Seth Austin Harding, Chien-Lung Lin, San-Tai Wang, and Shih-wei
Liao. Secured multi-layer blockchain framework for IoT aggregate verification.Advances in Sci-
ence, Technology and Engineering Systems Journal, 7(3):106–115, 2022. doi: 10.25046/aj070312.
[18] Chenxi Zhang, Rongxing Lu, Xiaodong Lin, Pin-Han Ho, and Xuemin Shen. An efficient identity-
based batch verification scheme for vehicular sensor networks. InIEEE INFOCOM 2008, pages
246–250, 2008.
[19] Cheng-Chi Lee and Yan-Ming Lai. Toward a secure batch verification with group testing for
V ANET.Wireless Networks, 19(6):1441–1449, 2013.
[20] Feng Tian. An agri-food supply chain traceability system for China based on RFID & blockchain
technology. In2016 13th International Conference on Service Systems and Service Management
(ICSSSM), pages 1–6, Piscataway, NJ, USA, 2016. IEEE.
[21] Feng Tian. A supply chain traceability system for food safety based on HACCP, blockchain &
Internet of Things. In2017 International Conference on Service Systems and Service Management,
pages 1–6, Piscataway, NJ, USA, 2017. IEEE.
[22] Jen-Hung Tseng, Yen-Chih Liao, Bin Chong, and Shih-wei Liao. Governance on the drug supply
chain via Gcoin blockchain.International Journal of Environmental Research and Public Health,
15(6):1055, 2018.
[23] Ali Dorri, Salil S. Kanhere, Raja Jurdak, and Praveen Gauravaram. Blockchain for IoT security
and privacy: The case study of a smart home. In2017 IEEE International Conference on Pervasive
Computing and Communications Workshops (PerCom Workshops), pages 618–623, 2017.
[24] Oscar Novo. Blockchain meets IoT: An architecture for scalable access management in IoT.IEEE
Internet of Things Journal, 5(2):1184–1195, 2018.
[25] Boohyung Lee and Jong-Hyouk Lee. Blockchain-based secure firmware update for embedded
devices in an Internet of Things environment.The Journal of Supercomputing, 73(3):1152–1167,
2017.
19

[26] Jen-Wei Hu, Lo-Yao Yeh, Shih-Wei Liao, and Chu-Sing Yang. Autonomous and malware-proof
blockchain-based firmware update platform with efficient batch verification for Internet of Things
devices.Computers & Security, 86:238–252, 2019.
[27] Louis Tremblay Thibault, Tom Sarry, and Abdelhakim Senhaji Hafid. Blockchain scaling using
rollups: A comprehensive survey.IEEE Access, 10:93039–93054, 2022.
[28] Dan Boneh, Ben Lynn, and Hovav Shacham. Short signatures from the Weil pairing. InAdvances in
Cryptology – ASIACRYPT 2001, volume 2248 ofLNCS, pages 514–532, Berlin, Heidelberg, 2001.
Springer.
[29] Sean Bowe. BLS12-381: New zk-SNARK elliptic curve construction. Electric Coin Company
blog,https://electriccoin.co/blog/new-snark-curve/, 2017.
[30] Dan Boneh, Craig Gentry, Ben Lynn, and Hovav Shacham. Aggregate and verifiably encrypted
signatures from bilinear maps. InAdvances in Cryptology – EUROCRYPT 2003, volume 2656 of
LNCS, pages 416–432, Berlin, Heidelberg, 2003. Springer.
[31] Dan Boneh, Manu Drijvers, and Gregory Neven. Compact multi-signatures for smaller blockchains.
InAdvances in Cryptology – ASIACRYPT 2018, volume 11273 ofLNCS, pages 435–464, Cham,
2018. Springer.
[32] Armando Faz-Hern ´andez, Sam Scott, Nick Sullivan, Riad S. Wahby, and Christopher A. Wood.
Hashing to elliptic curves. RFC 9380, Internet Research Task Force, 2023.
[33] Adi Shamir. Identity-based cryptosystems and signature schemes. InAdvances in Cryptology –
CRYPTO ’84, volume 196 ofLNCS, pages 47–53, Berlin, Heidelberg, 1985. Springer.
[34] Dan Boneh and Matt Franklin. Identity-based encryption from the Weil pairing. InAdvances in
Cryptology – CRYPTO 2001, volume 2139 ofLNCS, pages 213–229, Berlin, Heidelberg, 2001.
Springer.
[35] Jae Choon Cha and Jung Hee Cheon. An identity-based signature from gap Diffie-Hellman groups.
InPublic Key Cryptography – PKC 2003, volume 2567 ofLNCS, pages 18–30, Berlin, Heidelberg,
2003. Springer.
[36] Florian Hess. Efficient identity based signature schemes based on pairings. InSelected Areas
in Cryptography – SAC 2002, volume 2595 ofLNCS, pages 310–324, Berlin, Heidelberg, 2003.
Springer.
[37] Mihir Bellare, Juan A. Garay, and Tal Rabin. Fast batch verification for modular exponentiation and
digital signatures. InAdvances in Cryptology – EUROCRYPT ’98, volume 1403 ofLNCS, pages
236–250, Berlin, Heidelberg, 1998. Springer.
[38] Jan Camenisch, Susan Hohenberger, and Michael Østergaard Pedersen. Batch verification of short
signatures. InAdvances in Cryptology – EUROCRYPT 2007, volume 4515 ofLNCS, pages 246–
263, Berlin, Heidelberg, 2007. Springer.
20

[39] Anna Lisa Ferrara, Matthew Green, Susan Hohenberger, and Michael Østergaard Pedersen. Prac-
tical short signature batch verification. InTopics in Cryptology – CT-RSA 2009, volume 5473 of
LNCS, pages 309–324, Berlin, Heidelberg, 2009. Springer.
[40] Colin Boyd and Chris Pavlovski. Attacking and repairing batch verification schemes. InAdvances
in Cryptology – ASIACRYPT 2000, volume 1976 ofLNCS, pages 58–71, Berlin, Heidelberg, 2000.
Springer.
[41] Keisuke Hakuta, Yosuke Katoh, Hisayoshi Sato, and Tsuyoshi Takagi. Batch verification suitable
for efficiently verifying a limited number of signatures. InInformation Security and Cryptology –
ICISC 2012, volume 7839 ofLNCS, pages 425–440, Berlin, Heidelberg, 2013. Springer.
[42] EIP-197: Precompiled contracts for optimal ate pairing check on the elliptic curve alt bn128.
Ethereum Improvement Proposals,https://eips.ethereum.org/EIPS/eip-197, 2017.
[43] ConsenSys. GoQuorum documentation.https://docs.goquorum.consensys.io, 2023.
[44] Ralph C. Merkle. A digital signature based on a conventional encryption function. InAdvances
in Cryptology – CRYPTO ’87, volume 293 ofLNCS, pages 369–378, Berlin, Heidelberg, 1988.
Springer.
[45] Paul Miller. noble-curves: Audited and minimal JS implementation of elliptic curve cryptography.
https://github.com/paulmillr/noble-curves, 2025.
[46] Amos Fiat and Adi Shamir. How to prove yourself: Practical solutions to identification and signa-
ture problems. InAdvances in Cryptology – CRYPTO ’86, volume 263 ofLNCS, pages 186–194,
Berlin, Heidelberg, 1987. Springer.
[47] EthereumJS contributors. EthereumJS monorepo: VM, EVM and related libraries.https://
github.com/ethereumjs/ethereumjs-monorepo, 2025.
[48] Bitcoin Core contributors. libsecp256k1: Optimized C library for ECDSA signatures and
secret/public key operations on curve secp256k1.https://github.com/bitcoin-core/
secp256k1, 2025.
[49] arkworks contributors. arkworks: An ecosystem for zkSNARK programming.https://
arkworks.rs, 2022.
[50] EIP-4844: Shard blob transactions. Ethereum Improvement Proposals,https://eips.ethereum.
org/EIPS/eip-4844, 2024.
[51] EIP-7918: Blob base fee bounded by execution cost. Ethereum Improvement Proposals,https:
//eips.ethereum.org/EIPS/eip-7918, 2025.
[52] EIP-7594: PeerDAS – peer data availability sampling. Ethereum Improvement Proposals,https:
//eips.ethereum.org/EIPS/eip-7594, 2025.
[53] Jens Groth. On the size of pairing-based non-interactive arguments. InAdvances in Cryptology
– EUROCRYPT 2016, volume 9666 ofLNCS, pages 305–326, Berlin, Heidelberg, 2016. Springer.
doi: 10.1007/978-3-662-49896-5 11.
21

[54] EIP-1108: Reduce alt bn128 precompile gas costs. Ethereum Improvement Proposals,https:
//eips.ethereum.org/EIPS/eip-1108, 2019.
[55] Sattam S. Al-Riyami and Kenneth G. Paterson. Certificateless public key cryptography. InAd-
vances in Cryptology – ASIACRYPT 2003, volume 2894 ofLNCS, pages 452–473, Berlin, Heidel-
berg, 2003. Springer.
[56] Steven D. Galbraith, Kenneth G. Paterson, and Nigel P. Smart. Pairings for cryptographers.Discrete
Applied Mathematics, 156(16):3113–3121, 2008.
[57] David Pointcheval and Jacques Stern. Security arguments for digital signatures and blind signatures.
Journal of Cryptology, 13(3):361–396, 2000.
22

论文原文与出版信息

arXiv
查看论文原文

继续探索