Research on Data Heterogeneous Robust Federated Learning with Privacy Protection in Internet of Things

YANG Zhigang; WANG Zhuotong; WU Dapeng; WANG Ruyan; WU Yu; LÜ Yi

doi:10.11999/JEIT221193

Volume 45 Issue 12

Dec. 2023

Turn off MathJax

Article Contents

Article Navigation > Journal of Electronics & Information Technology > 2023 > 45(12): 4235-4244

Zou Chuanyun, Ao Faliang, Huang Xiangfu . ANALYSIS OF OPTICAL PPM CHANNEL CAPACITY WITHOUT BACKGROUND NOISE[J]. Journal of Electronics & Information Technology, 2000, 22(4): 682-686.

Citation:

YANG Zhigang, WANG Zhuotong, WU Dapeng, WANG Ruyan, WU Yu, LÜ Yi. Research on Data Heterogeneous Robust Federated Learning with Privacy Protection in Internet of Things[J]. Journal of Electronics & Information Technology, 2023, 45(12): 4235-4244. doi: 10.11999/JEIT221193

Zou Chuanyun, Ao Faliang, Huang Xiangfu . ANALYSIS OF OPTICAL PPM CHANNEL CAPACITY WITHOUT BACKGROUND NOISE[J]. Journal of Electronics & Information Technology, 2000, 22(4): 682-686.

Citation:

PDF( 8464 KB)

Research on Data Heterogeneous Robust Federated Learning with Privacy Protection in Internet of Things

doi: 10.11999/JEIT221193

YANG Zhigang^{1, 2, 3, 4},
WANG Zhuotong^{2, 3, 4},
WU Dapeng^{2, 3, 4
,
,},
WANG Ruyan^{2, 3, 4},
WU Yu¹,
LÜ Yi^{2, 3, 4}

1.
School of Cyber Security and Information Law, Chongqing University of Posts and Telecommunications, Chongqing 400065, China
2.
School of Communication and Information Engineering, Chongqing University of Posts and Telecommunications, Chongqing 400065, China
3.
Advanced Network and Intelligent Connection Technology Key Laboratory of Chongqing Education Commission of China, Chongqing 400065, China
4.
Chongqing Key Laboratory of Ubiquitous Sensing and Networking, Chongqing 400065, China

Funds: The National Natural Science Foundation of China (61901071, 61871062, 61771082, 62271096, U20A20157), The Natural Science Foundation of Chongqing (cstc2020jcyj-zdxmX0024), The University Innovation Research Group of Chongqing Foundation (CXQT20017), The Program for Innovation Team Building at Institutions of Higher Education in Chongqing (CXTDX201601020), The Youth Innovation Group Support Program of ICE Discipline of Chongqing University of Posts and Telecommunications (SCIE-QN-2022-04)

Received Date: 2022-09-14
Rev Recd Date: 2023-01-15

Available Online: 2023-02-08

Publish Date: 2023-12-26

Abstract

Abstract

Federated learning allows the effective flow of data value without leaving the local data, which is considered to be an effective way to balance data sharing and privacy protection in the Internet of Things (IoT) scenario. However, federated learning systems are vulnerable to Byzantine attacks and inference attacks, leading to the robustness of the system and the privacy of the data being compromised. The data heterogeneity and resource bottlenecks of IoT devices also pose significant challenges to the design of privacy-preserving and Byzantine-robust algorithms. In this paper, a data resampling of robust aggregation method Re-Sim applicable to heterogeneous IoT is proposed, which achieves robust aggregation by measuring directional similarity and normalized update magnitude, and uses data resampling to enhance robustness in data heterogeneous environments. Meanwhile, a Lightweight Security Aggregation (LSA) protocol is proposed to ensure data privacy while taking into account model robustness, accuracy and computational overhead, and the privacy of the protocol is theoretically analyzed. Simulation results show that the proposed scheme can effectively resist Byzantine attacks and inference attacks in the case of data heterogeneity. The proposed scheme improves the accuracy by 1%～3% compared to the baseline method, while reducing the client-side computational overhead by 79%.
- Internet of Things (IoT),
- Federated learning,
- Robust aggregation,
- Secret sharing

FullText(HTML)

1. 引言

无线通信领域每一次技术更新都伴随着带宽的拓展，以支持更大的数据传输速率和数据容量。射频功率放大器(功放)作为无线通信系统中最重要的有源模块，其性能的好坏直接影响着无线通信系统性能的通信质量。因此宽带高效率功率放大器的设计仍然是当今研究的热点。而宽带功率放大器设计的难点在于宽带范围内功放的最优阻抗随频率会有较大的变化，关键在于如何选取一个最优阻抗解使得其在宽带范围内得到最优阻抗以及宽带匹配网络的设计。而传统的A, AB, B, C类功放很难在宽带范围内同时实现高效率和高输出功率。而D, E类等开关类功放在GHz以上频段会发生波形失真，且受到功放器件的开关限制，很难有实际用途^[1,2]。F类功放则需要对高次谐波进行精确的控制，在宽带范围内很难实现。由Cripps等人^[3]在2009年提出一种新型谐波控制类功放即J类功放，它相比于F类功放二次谐波阻抗不再需要严格的开路或者短路，使得J类功放具有实现宽带的潜力。而在J类功放的基础上提出的连续功率放大器，相比于对各次谐波和基波阻抗进行精确控制的传统谐波控制类型，连续型功放模式有着较为丰富的最佳阻抗解空间，简化宽带匹配网络的设计复杂度^[4,5]。而混合连续类功放在包含连续F类与连续B/J类的基础上将阻抗解空间进一步扩大，相比于传统连续类它实现宽带功放的潜力更大。

本文首先分析混合连续类的漏极电压公式进而推导出阻抗设计空间，然后通过使用一种新型的谐波网络，再采用阶跃阻抗低通滤波结构作为输出匹配网络，最后基于功率密度大、击穿电压高、电子饱和漂移速度高的第3代半导体GaN HEMT^[6,7]设计了一款性能优越的超宽带功放。

2. 混合连续类功放原理

最早由英国卡迪夫大学的Cripps等人^[3]提出的F,J类功放通过降低导通角来提高功放的效率。通过假定电流谐波分量在器件输出端为短路状态，保证了功放输出电压波形为正弦波^[8]。在Cripps等人推导过程中定义归一化电压公式为

$v(\theta ) = 1 - {v_{1r}}\cos \theta + {v_{1q}} + \sum\limits_n {{v_{nq}}} \sin (n\theta )$

(1)

随着 ${v_{1r}}$ , ${v_{1q}}$ , ···, ${v_{nq}}$ 的取值不同，漏极电压表现出不同的时域波形。为了保证良好的线性防止功放漏压进入膝值区，因膝值区电流急剧下降会带来削波、增益压缩、AM-PM失真等一系列不良的影响。Cripps定义了“Zero-Grazing”: $v\left( \theta \right) \ge 0$ 。Crippls在Rhodes和Raab的基础上提出了J类功放也满足 $v(\theta ) = 0$ , ${v'}(\theta ) = 0$ 同时为0的特殊解，此时2次谐波为纯容性，基波为包含了感性电抗成分的复阻抗，并且能够得到与B类相等的输出功率和漏极效率。于是连续B/J类功放归一化电压波形可描述为

$v\left( \theta \right){\rm{ = }}\left( {1 - \cos \theta } \right) \cdot \left( {1 - \alpha \sin \theta } \right), - 1 < \alpha < 1$

(2)

当 $\alpha = 0$ 时电压波形为B类波形，当 $\alpha = 1$ 时此时为J类功放，当 $\alpha$ 在–1～1变化时对应一系列的B类和J类的功放类型。与连续B/J类的来源相似对F类的推广得到了连续F类的功放模型其电压波形表达式为^[8]

${v_{{\rm{CF}}}} = {\left(1 - \frac{2}{{\sqrt 3 }}\cos \theta \right)^2} \cdot \left(1 + \frac{1}{{\sqrt 3 }}\cos \theta \right) \cdot \left(1 - \beta \sin \theta \right)$

(3)

其中， $\beta$ 值在–1～1变化。结合连续F类和B/J类的电压波形，Chen等人^[9]推导出了混合连续类功放漏极电压表达式

$\begin{split} {v_{{\rm{DS}}}}(\theta ) =& (1 - {\alpha '}\cos \theta + {\beta '}\cos 3\theta ) \\ & \cdot (1 - \gamma \sin \theta ),\; { - 1 < \gamma < 1} \end{split}$

(4)

当 ${\alpha '} = {2 / {\sqrt 3 }},{\beta '} = {1 / {3\sqrt 3 }}$ 时为连续F类功放的漏极电压表达式。当 ${\alpha '} = 1,{\beta '} = 0$ 为连续B/J类功放的漏极电压表达式。当 ${\alpha '}$ , ${\beta '}$ 取其他值时也可对应其他类型的功放，图1展示不同 ${\alpha '}$ , ${\beta '}$ , $\gamma$ 时对应的漏极电压波形。

图 1 混合连续类归一化漏极电压波形

下载: 全尺寸图片幻灯片

根据电压电流表达式可以推导出它在电流源平面下的基波阻抗及2次谐波阻抗

$\begin{split} & {Z_1}{\rm{ = }}{R_{{\rm{opt}}}} \cdot ({\alpha '} + {\rm{j}}\gamma ) \\ & {Z_2} = - \frac{{3\pi }}{8}{R_{{\rm{opt}}}} \cdot \gamma \cdot ({\alpha '} + {\beta '}) \\ \end{split}$

(5)

其中， ${R_{{\rm{opt}}}} = {{2({V_{{\rm{DC}}}} - {V_{{\rm{knee}}}})} / {{I_{{\rm{Max}}}}}}$ 通过Load-pull技术得到 ${R_{{\rm{opt}}}} = 36\;\Omega$ , ${V_{{\rm{DC}}}}$ 是漏极电压， ${I_{{\rm{Max}}}}$ 为晶体管所能承受的最大电流。当 ${\alpha '}$ , ${\beta '}$ , $\gamma$ 取不同值时得到电流源平面的阻抗空间如图2所示。

图 2 混合连续类电流源平面的阻抗变化轨迹

下载: 全尺寸图片幻灯片

3. 混合连续类功放的电路设计

3.1 基于波形工程的电路设计

“波形工程”的主要思想在于通过合理的增加谐波分量来调控电压和电流波形使得功率管漏极输出电压和电流发生较小的重叠，从而减小漏极功率消耗，同时保证较好的谐波抑制，从而在较宽的频带范围内维持高效率^[10,11]。在新型连续性的拓展中，在保证某一固定偏置下，通过对谐波以及基波阻抗成分进行调整，控制晶体管的漏极电流与电压之间的重叠尽可能减小，使得该连续型模型在一个更为宽泛的阻抗解空间中保持合适的输出功率以及效率。

图3是本文所采用的谐波控制网络，该谐波控制网络是基于中心频率IB(Impedance Buffer)^[12,13]理论所设计的。通过将1/4波长的开路短截线与半波长短路短截线并联实现对功放谐波控制。其中 ${\theta _{S1}}^{3{f_0}} = {90^\circ }$ , ${Z_{S1}}$ 为自由设计值，因为在平面 ${P_1}$ 引入了 ${\rm{I}}{{\rm{B}}_1}$ 则在3次谐波为短路。传输线 ${L_1}$ 的作用主要是在 ${\rm{I}}{{\rm{B}}_1}$ 之前3次谐波阻抗虚部部分不受 ${\rm{I}}{{\rm{B}}_1}$ 之后网络影响。特征阻抗 ${Z_{L1}}$ 和传输线 ${L_1}$ 的电长度 ${\theta _{L1}}$ 满足关系式

图 3 谐波调谐网络

下载: 全尺寸图片幻灯片

${Z_0}(3{f_0}) = {\rm{j}} \cdot {Z_{L1}} \cdot \tan ({\theta _{L1}}^{3{f_0}})$

(6)

其中， ${Z_0}(3{f_0})$ 是通过负载牵引技术在 $3{f_0}$ 得到的最优阻抗， ${Z_{L1}}$ 为自由设计空间。而传输线 ${L_1}$ 的电长度 ${\theta _{L1}}$ 求解如

${\theta _{L1}}^{3{f_0}} = \arctan \left(\frac{{{Z_0}(3{f_0})}}{{{\rm{j}}{Z_{L1}}}}\right)$

(7)

通过在 $2{f_0}$ 半波长短路短截线结合传输线 ${L_2}$ ，采用同样的方式可以在 $2{f_0}$ 处获得最优谐波阻抗。其中 ${\theta _{S2}}^{2{f_0}} = 180^\circ$ , ${Z_{S2}}$ 为自由设计值。由于在 $2{f_0}$ 的短路条件需要满足平面 ${P_2}$ ，传输线 ${L_2}$ 在 $2{f_0}$ 的电长度满足关系

${\theta }_{L2}{}^{2{f}_{0}}=\mathrm{arctan}\left(\frac{{Z}'_{1}(2{f}_{0})}{{\rm{j}}{Z}_{L2}}\right)$

(8)

而平面 ${P_1}'$ 处的阻抗 $Z'(2{f_0})$ 为

$Z'(2{f_0}) = \frac{{{\rm{j}}{Z_1}(2{f_0}){Z_{S1}}\cot \left(\dfrac{\pi }{2}.\dfrac{{2{f_0}}}{{3{f_0}}}\right)}}{{2{Z_1}(2{f_0}) + {\rm{j}}{Z_{S1}}\cot \left(\dfrac{\pi }{2}.\dfrac{{2{f_0}}}{{3{f_0}}}\right)}}$

(9)

在平面 ${P_1}$ 处的阻抗 ${Z_1}(2{f_0})$ 可通过式(10)计算得到

${Z_1}(2{f_0}) = {Z_{L1}} \cdot \frac{{{Z_0}(2{f_0}) - {\rm{j}}{Z_{L1}}\tan \left({\theta _{S1}}^{3{f_0}} \cdot \dfrac{{2{f_0}}}{{3{f_0}}}\right)}}{{{Z_{L1}} - {\rm{j}}{Z_0}(2{f_0})\tan \left({\theta _{S1}}^{3{f_0}} \cdot \dfrac{{2{f_0}}}{{3{f_0}}}\right)}}$

(10)

其中， ${Z_0}(2{f_0})$ 可通过负载牵引技术得到。因此谐波网络的所有参数可由以上的式(6)—式(10)确定。

3.2 匹配网络的设计

由于谐波网络已经将2次谐波阻抗控制在高效率和高输出功率区域并且尽量维持在Smith边缘，因此输出匹配网络主要完成对基波阻抗匹配。为了实现对基波匹配网络的精确设计，在大信号模型下采用负载牵引技术得到最优基波阻抗。并采用最平坦低通原型结构^[14]，利用阶梯阻抗变换线^[15,16]将最优负载阻抗匹配到 $50\;\Omega$ 。利用ADS中的随机优化算法得到的最终结构如图4所示。图5给出了该功放在其工作带宽内的阻抗分布曲线，可以看出其阻抗分布与理论推导一致。较好地实现了对功放基波阻抗的匹配和对高次谐波的抑制。输入匹配网络主要是通过负载牵引技术得到最优源阻抗并通过Smith圆图进行匹配，这里不再详细说明了，具体的参数将在后面给出。

图 4 优化后的电流源平面的谐波网络加匹配结构

下载: 全尺寸图片幻灯片

图 5 在电流源平面的输出网络阻抗变换轨迹

下载: 全尺寸图片幻灯片

4. 整体电路仿真和测试

4.1 整体电路仿真

本次设计采用的晶体管为Cree公司的CGH40010F GaN HEMT的封装功放管，漏极电压为28 V，栅极电压为–2.8 V，采用Rogers4003C的介质基板，介电常数3.38厚度为0.813 mm。整体电路的结构如图6所示。

图 6 整体电路结构图

下载: 全尺寸图片幻灯片

图7给出了在大信号下ADS仿真电压电流波形，可以看出电压电流波形基本上满足连续F类与连续B/J类的电压电流波形形式，并且电压幅度接近 $2{V_{{\rm{DD}}}}$ 。

图 7 不同频点下大信号仿真电压电流波形

下载: 全尺寸图片幻灯片

4.2 功放实物与实际测试

采用Rogers 4003c板材加工的功放PCB实物照片如图8所示。进行大信号测试功放的输出功率、漏极效率、功率附加效率、增益和仿真的结果如图9所示。最终在0.3～3.5 GHz频段内得到的增益为10～18 dB，漏极效率为58.4%～72.6%，输出功率为39.8～41.2 dBm。表1是与近些年来国内外相关功放设计的效果对比。

图 8 混合连续类功放实物图

下载: 全尺寸图片幻灯片

表 1 本文设计功放与国内外相关功放性能对比

文献对比	工作模式	频率(GHz)	BW(%)	PAE(%)	DE(%)	增益(dB)	Pout(dBm)	年份
文献[2]	F类/逆F类	2.4～4.2	54.5	–	55.0～82.0	11～13.6	39.5～41.9	2019
文献[11]	连续类	0.8～3.6	127.0	–	55.8～74.1	10.2～12.2	39.5～42.1	2016
文献[13]	连续F类	0.5～2.3	128.5	52.7～80.7	60.0～81.0	11.7～25.3	39.2～41.2	2017
文献[15]	BJ类/F-1类	2.2～2.8	24.0	58.0～75.0	65.9～79.7	12.0～18.0	41.0～43.0	2019
文献[16]	连续B/J类	1.2～2.6	74.0	–	63.0～68.0	10.0～13.6	40.5～41.6	2018
本文	混合连续	0.3～3.5	168.4	–	58.4～72.6	10.0～18.0	39.8～41.2	2019

下载: 导出CSV

| 显示表格

图 9 仿真与实测对比图

下载: 全尺寸图片幻灯片

5. 结束语

本文通过分析连续F类、B/J类功放电压波形确定混合连续类功放的阻抗设计空间，基于阻抗缓冲(IB)概论提出了一种设计谐波网络的方法，有效地将基波、2次谐波和3次谐波的阻抗控制到理论提出的设计空间内。本文采用Cree公司的CGH40010F GaN HEMT的功放管，设计一款宽带跨3个倍频程的混合连续类功放。在实际测量中在0.3～3.5 GHz内有58.4%以上的漏极效率，输出功率有39.8 dBm以上。

References(20)

References

[1]	黄新林, 郑人华. 基于强化学习的802.11ax上行链路调度算法[J]. 电子与信息学报, 2022, 44(5): 1800–1808. doi: 10.11999/JEIT210590 HUANG Xinlin and ZHENG Renhua. 802.11ax uplink scheduling algorithm based on reinforcement learning[J]. Journal of Electronics &Information Technology, 2022, 44(5): 1800–1808. doi: 10.11999/JEIT210590
[2]	MCMAHAN B, MOORE E, RAMAGE D, et al. Communication-efficient learning of deep networks from decentralized data[C]. The 20th International Conference on Artificial Intelligence and Statistics, Fort Lauderdale, USA, 2017: 1273–1282.
[3]	BLANCHARD P, EL MHAMDI E M, GUERRAOUI R, et al. Machine learning with adversaries: Byzantine tolerant gradient descent[C]. The 31st International Conference on Neural Information Processing Systems, Long Beach, USA, 2017: 118–128.
[4]	YIN Dong, CHEN Yudong, RAMCHANDRAN K, et al. Byzantine-robust distributed learning: Towards optimal statistical rates[C]. The 35th International Conference on Machine Learning, Stockholm, Sweden, 2018: 5636–5645.
[5]	MELIS L, SONG Congzheng, DE CRISTOFARO E, et al. Exploiting unintended feature leakage in collaborative learning[C]. 2019 IEEE Symposium on Security and Privacy (SP), San Francisco, USA, 2019: 691–706.
[6]	XIONG Jinbo, BI Renwan, TIAN Youliang, et al. Toward lightweight, privacy-preserving cooperative object classification for connected autonomous vehicles[J]. IEEE Internet of Things Journal, 2022, 9(4): 2787–2801. doi: 10.1109/JIOT.2021.3093573
[7]	BI Renwan, XIONG Jinbo, TIAN Youliang, et al. Achieving lightweight and privacy-preserving object detection for connected autonomous vehicles[J]. IEEE Internet of Things Journal, 2023, 10(3): 2314–2329. doi: 10.1109/JIOT.2022.3212464
[8]	ZAWAD S, ALI A, CHEN Pinyu, et al. Curse or redemption? How data heterogeneity affects the robustness of federated learning[C]. The 35th AAAI Conference on Artificial Intelligence, Vancouver, Canada, 2021: 10807–10814.
[9]	ZHAI Kun, REN Qiang, WANG Junli, et al. Byzantine-robust federated learning via credibility assessment on non-IID data[J]. arXiv: 2109.02396, 2021.
[10]	LI Liping, XU Wei, CHEN Tianyi, et al. RSA: Byzantine-robust stochastic aggregation methods for distributed learning from heterogeneous datasets[C]. The 33rd AAAI Conference on Artificial Intelligence, Hawaii, USA, 2019: 1544–1551.
[11]	ZHANG Chengliang, LI Suyi, XIA Junzhe, et al. BatchCrypt: Efficient homomorphic encryption for cross-silo federated learning[C]. The 2020 USENIX Conference on Usenix Annual Technical Conference, Boston, USA, 2020: 33.
[12]	FU Anmin, ZHANG Xianglong, XIONG Naixue, et al. VFL: A verifiable federated learning with privacy-preserving for big data in industrial IoT[J]. IEEE Transactions on Industrial Informatics, 2022, 18(5): 3316–3326. doi: 10.1109/TII.2020.3036166
[13]	SO J, GÜLER B, and AVESTIMEHR A S. Byzantine-resilient secure federated learning[J]. IEEE Journal on Selected Areas in Communications, 2021, 39(7): 2168–2181. doi: 10.1109/JSAC.2020.3041404
[14]	LIU Xiaoyuan, LI Hongwei, XU Guowen, et al. Privacy-enhanced federated learning against poisoning adversaries[J]. IEEE Transactions on Information Forensics and Security, 2021, 16: 4574–4588. doi: 10.1109/TIFS.2021.3108434
[15]	HSIEH K, HARLAP A, VIJAYKUMAR N, et al. Gaia: Geo-distributed machine learning approaching LAN speeds[C]. The 14th USENIX Conference on Networked Systems Design and Implementation, Boston, USA, 2017: 629–647.
[16]	LI Yiran, LI Hongwei, XU Guowen, et al. Efficient privacy-preserving federated learning with unreliable users[J]. IEEE Internet of Things Journal, 2021, 9(13): 11590–11603. doi: 10.1109/JIOT.2021.3130115
[17]	ZHU Wanchuang, ZHAO B Z H, LUO S, et al. MANDERA: Malicious node detection in federated learning via ranking[J]. arXiv: 2110.11736, 2021.
[18]	LECUN Y, BOTTOU L, BENGIO Y, et al. Gradient-based learning applied to document recognition[J]. Proceedings of the IEEE, 1998, 86(11): 2278–2324. doi: 10.1109/5.726791
[19]	XIAO Han, RASUL K, and VOLLGRAF R. Fashion-MNIST: A novel image dataset for benchmarking machine learning algorithms[J]. arXiv: 1708.07747, 2017.
[20]	WANG Hao, KAPLAN Z, NIU Di, et al. Optimizing federated learning on non-IID data with reinforcement learning[C]. IEEE INFOCOM 2020 - IEEE Conference on Computer Communications, Toronto, Canada, 2020: 1698–1707.

Relative Articles

Supplements(0)

Cited By

Cited by
Periodical cited type(0)
Other cited types(3)

Proportional views

Proportional views

通讯作者: 陈斌, bchen63@163.com

1.
沈阳化工大学材料科学与工程学院沈阳 110142

Figures(8) / Tables(1)

Get Citation

PDF

XML

Article Metrics

Article views (869) PDF downloads(251)