摘要

借助無線通信標準，例如LTE和5G，對較高的數據速率和光譜效率的重點促使無線原始設備制造商（OEM）采用新的傳輸格式，例如正交頻施加多元化（OFDM）。但是，這些信號在其信封中具有較大的波動，由于其高峰值功率比（PAPR）而尤其容易受到非線性功率放大器（PA）畸變的影響。使用該高PAPR信號，PA非線性可以產生大量的信號扭曲，從而增加了位錯誤率（BER）并因此而降低了信噪比。本文回顧了PAPR，它們起源于其中，如何分解發射陣容的RF組件，以及如何擺脫它們或至少減輕其對信號鏈的影響。

介紹

較新的調制格式（例如OFDM）和各種形式的正交振幅調制（QAM）在信號信封中具有較大的波動。這會在信號中產生高papr。在非線性PA上播放高PAPR信號會產生光譜再生。光譜再生是指由增益壓縮引起的新頻率，而不是在原始輸入中。高papr會導致帶內變形，從而降低了整個系統的BER性能。我們將討論一種解決方案，以幫助找到使用數字預性（DPD）和降低峰值因子（CFR）發動機之間的效率和線性之間的正確系統權衡。

OFDM調制 - 每個人都在這樣做！

在LTE和5G系統中，正在并行傳輸幾個載體的載體聚集用于提高帶寬和數據速率。這些網絡杠桿OFDM調制是一種非常熟練且使用廣泛使用的多載波傳輸技術，可以提高光譜效率并降低多路徑反射對接收器解碼信號的能力的影響。使用OFDM，最終波形是攜帶信息的子載波的正交總和，每個子載波都有其自己的中心頻率和調制方案。在時間域中，有時這些子載波的峰可以對齊以產生一個巨大的OFDM波形峰。OFDM的一個獨特特征是，子載波波形是正交組合的，因此一個子載波的空null（或零振幅）與圖1所示的其他子載波的峰值相吻合。這提供了通道帶寬相對有效的使用，從而與傳統的單一單位模塊相比，相對有效地使用了該通道帶寬。

圖1 DM子載體波形的多載體。

OFDM還有其他幾個好處，包括其對多徑褪色的穩健性。但是，OFDM調制的主要問題之一是傳輸波形遭受高papr的影響。圖2顯示了各種普通移動技術或調制類型的PAPR。可以觀察到，隨著新標準或調制技術的出現，典型的PAPR一直在穩步增加。

圖2。各種調制技術的典型PAPR。

ofm信號

如前所述，由OFDM調制啟用的載波聚集用于提高5G系統中的帶寬和數據速率。 OFDM還會產生信號的信號，其信封是非構恒定的，這可能會導致較高的PAPR，這可能會導致系統損壞。如果RF信號系列（尤其是PA）中的RF功率組件不能適當地指定以處理預期的電壓峰，則這些組件可能會失敗。一個大的PAPR通過將PA的效率推向飽和，從而降低了PA的效率，即其非線性工作區域，導致失真導致信號的光譜擴散。對于非穩定 - 內玻璃數字調制方案而言，PA的線性一直是一個關鍵的設計問題。圖3顯示了一個時域LTE 64-QAM信號，該信號在ADRV9040的發射器輸出。

圖3。造成大峰的子載波的正交求和的例證。

互補分布函數

由于其形式，OFDM信號需要采用統計方法進行適當的測量。補充累積分布函數（CCDF）用于評估RF信號鏈中的PAPR降低性能。圖4a顯示了LTE下行鏈路10 MHz帶寬和64個QAM子載波調制信號的傳輸波形。圖4b中的CCDF顯示，信號功率在0.01％的時間內至少超過7.4 dB。理論最大峰值出現在0％的概率上，該概率在此圖上不確定。該痕量在約7.4 dB的PAPR處與X軸（0.01％或概率為10E ^-4^ ）相交。這表明每10,000個中的一個樣本預計將超過7.4 dB的平均功率。

圖4。具有10 MHz帶寬和64個QAM子載波調制的LTE下行鏈路的CCDF。

仔細觀察CCDF圖后，觀察Y軸是累積的概率，通常在對數刻度上繪制。 X軸是DB中繪制的功率。該圖顯示信號功率超過平均功率或高于平均功率的概率或時間百分比。從本質上講，對于每個功率水平，CCDF圖描述了信號在平均功率水平以上的時間。隨著CCDF曲線向右移動，我們的峰值功率與平均功率的比率增加。

CCDF圖可驗證線性操作，并且在PA之后更常見。與通常在不同功率水平下增益變化的方法相比，它可以更準確地描述信號壓縮。波峰因子發生的統計分析使其成為設計人員評估放大器壓縮對系統BER和/或誤差矢量幅度（EVM）的影響的寶貴工具。

那么，為什么PAPR很重要？

PAS本質上是非線性的，在線性和效率之間表現出權衡。常見的非線性問題是增益壓縮和相失真，包括帶外和帶外變形。這些因素中的每一個都會降低系統的性能，并創建帶外光譜再生，從而導致相鄰的渠道干擾，并違反了監管機構規定的帶外排放標準。

在測試PA時，輸入幅度逐漸增加，直到測得的比率降低1 dB，代表1 dB增益壓縮。 1 dB壓縮點是功績的關鍵圖形，它為RF設計師提供了有關其放大器性能的假設。從本質上講，放大器的1 dB壓縮點被定義為從其小信號值降低設備增益下降1 dB的輸出功率。該參數通常用作放大器非線性開始的參考點，大約等于放大器的最大可用峰輸出功率。這就是為什么許多RF設計人員通常估計其PA的最大工作輸出功率比其1 dB壓縮點低幾dB的原因。這使得在PA的1 dB壓縮點成為關鍵練習，為了使PAPR高的信號永遠不會飽和PA。 PAPR的另一個名稱是波峰因素。圖5顯示了所示的1 dB壓縮點的AM-AM曲線。

圖5。具有1 dB壓縮點的AM-AM曲線。

現在，設計師已經評估了PA并確定了其1 dB的壓縮點，他們需要在其線性區域中以輸入功率向后進行操作PA（例如，在其工作曲線的線性部分內以較低的功率操作PA）以避免光譜增長，對嗎？好吧，不完全是！

僅僅備份了遠離PA的飽和點的輸入，肯定可以幫助避免討論的所有非線性問題，但會導致非常低的效率并增加散熱耗散。通過增加系統的功耗來解決這種低效率問題并不是可行的權衡。如圖2所示，隨著標準機構具有新的調制方案的創新性，可以更好地利用現有頻譜，這導致信號具有越來越高的波峰因素。因此，從長遠來看，使用PA返回實施策略將不起作用。本文的下一個部分將討論兩種實施策略，結合在一起，將使PA運行到其飽和點，同時仍保持良好的線性性并大大提高其效率。第一個使用振幅剪切技術來減少PAPR，第二種方法是將PA在其預期功率范圍內的非線性響應線性化。

非常成功的數字前端解決方案的兩個功能

在無線數字前端（DFE）系統中，涵蓋了廣泛的子系統，包括DPD，Digital Upconversion（DUC），數字下降轉換（DDC）和CFR。還有其他重要領域，例如DC偏移校準，脈搏形成，圖像排斥，數字混合，延遲/增益/不平衡補償，誤差校正和其他相關塊。 DPD電路使用PA輸出處捕獲的數據來線性化PA輸出。 DPD通過允許PAS更有效地操作來改善系統線性，而CFR有助于限制信號PAPR。使用CFR減少信號的動態范圍后，使用DPD發動機，并允許PA在線性區域上方操作。盡管這些塊中的每個塊涵蓋了DFE的關鍵功能，但本文的這一部分將僅關注CFR和DPD塊。

波峰因子降低

OFDM波形的大多數輸入信號將在PA的線性范圍內。但是，如前所述，該信號的峰值可能超過PA的線性工作范圍，由于其對系統損害的貢獻，它引起了長期可靠性問題。同樣，非常需要在不飽和的情況下以最高輸入功率驅動PA。為了避免由于峰值而導致的飽和度，使用了CFR，而不是衰減整個信號，而是減弱了PA線性范圍以上信號的部分。簡而言之，CFR有助于保持PA線性。

抑制峰時，這會導致恒定的輸出功率，從而確保信號保持在PA的線性范圍內。請注意，CFR不是線性化技術，而是一種效率提高方案。 CFR憑借其最有效的實施，消除了發射信號的峰值，以降低峰與平均比率，同時遵守所需的光譜發射掩碼，相鄰的通道功率比和EVM規格。圖6顯示了檢測到的信號峰高于閾值水平。峰的幅度降低至低于某些目標值。這通常是過濾以重塑信號頻譜。

圖6。檢測到的信號峰高于閾值。

CFR的缺點是剪輯會導致帶內信號失真，導致BER性能下降和帶外輻射，從而將帶外干擾信號施加到相鄰的通道上。簡而言之，剪輯的后果是信號ACLR和EVM差。過濾夾板信號通常用于以峰值再生成本減少帶外輻射。

數字預性

使用DPD，可以將PA推動以直至飽和區域，而不會損害其線性特性。 DPD允許RF設計人員在PA的有效但非線性區域操作其系統，同時保留OFDM調制所需的發送信號線性。換句話說，使用DPD，PA的線性區域擴展了。 DPD發動機通過對PA的逆AM-AM和AM-PM特性進行建模來產生預性系數。本質上，DPD專注于提高PA在其峰值效率點運行時產生的信號的質量。 DPD旨在引入彌補PA增益的反非線性。這是一種通過將精確的反抗原引入輸入波形，以補償PA的帶內非線性產品的輸入波形，以提高非線性PA的線性。圖7顯示了DPD線性化PA響應的概念。

圖7。用于線性化PA響應的DPD的通用概念：（a）典型的AM-AM曲線顯示整個線性區域為綠色； （b）DPD的基本概念及其如何提高功率放大器效率。

它根據預測數字域中的傳輸數據的原理來取消模擬域中的PA壓縮引起的失真。 DPD的方法可以從簡單的解決方案（例如基本查找表（LUT））到更復雜的實時信號處理方法。 DPD實現可以分類為具有內存的無內存模型和模型。

無內存的DPD

無內存的DPD僅基于當前樣本來糾正智商樣本的幅度和相位。嚴格無記憶的PA可以以AM-AM和AM-PM轉換為特征。這種瞬時非線性通常以PA的AM-AM和AM-PM響應為特征，其中PA輸出的輸出信號振幅和相位偏差作為其當前輸入幅度的函數。因此，可以通過其AM-AM和AM-PM響應來表征無內存的PA。這些測量用于創建將每個輸入功率/相組合與產生所需線性輸出所需的功率/相相關聯的LUT數據。無內存DPD的優點是它可以作為查找表實現相對簡單。圖7a和7b顯示了具有和沒有DPD校正的PA的AM-AM響應，應用于2×100 MHz，400 MHz帶寬4096樣品數據集。

圖8。在2×100 MHz，400 MHz帶寬信號上具有和不帶DPD的PA的AM-AM響應。

帶有內存的DPD

隨著傳輸信號帶寬的變化，PA將開始表現出記憶效應。這些是某些組件中的頻率響應，例如偏置網絡，解耦電容器，電源電路，或可以歸因于活動設備的熱常數。結果，PA的當前輸出不僅取決于當前輸入，還取決于過去的輸入值。在這種情況下，這意味著PA已成為具有內存的非線性系統。帶有內存的DPD基于以前的幾個樣本及其相互依賴性，糾正了iQ數據樣本的振幅和階段。 PA的響應通常不僅取決于當前信號振幅，還取決于先前樣品的幅度。因此，數字預性也需要具有內存結構，這是DPD的數學骨干。 Volterra系列是具有內存的非線性最通用的多項式類型，用于模擬具有內存的非線性系統。因此，引入內存的最通用方法是使用Volterra系列。有關與Volterra系列的建模PA失真背后的數學細節的精心討論，該系列超出了本文的范圍，請參閱Masterson（2022）。這超出了本文的范圍，請參考Masterson（2022）。這超出了本文的范圍，請參考Masterson（2022）。^2^

輕松設計5G RF信號鏈的框架

ADRV9040 RF收發器提供了一個簡化的框架，可輕松設計，實現和測試5G通信系統的RF信號鏈陣容。一個離散的大型MIMO系統需要以其離散的部署形式進行四個芯片級別，包括RF收發器，DFE FPGA，基帶FPGA/ASIC和控制FPGA。由于該收發器附帶了DFE集成，因此它消除了對競爭性離散解決方案中使用的幾種FPGA的需求，在該解決方案中，在計算機代碼中實現了DPD，CFR，DUC和DDC塊。 FPGA的實施通常是昂貴的，并且饑餓。這款高度集成的RF收發器有助于消除這種渴望強力的專用FPGA。在本文的這一部分中，我們強調了該RF收發器和提議的框架，用于檢查典型的PA增益陣容并通過在設備內實現寄存器寫入噪聲限制的理智檢查。

芯片（SOC）上的這種高度集成的射頻敏捷收發器系統提供了八個發射器，兩個用于監視發射機通道的觀察接收機，八個接收器，集成的本地振蕩器（LO）和時鐘合成器以及數字信號處理功能，以提供具有重要數字前端能力的完整收集器。該設備提供了由小細胞無線電單元（RUS），宏觀4G/5G RUS和大量MIMO RUS所需的高無線電性能和低功耗。完整的收發器子系統包括自動和手動衰減控制，DC偏移校正，正交誤差校正和數字過濾。收發器具有完全集成的數字前端，該數字前端支持一些關鍵塊，包括DPD（最多400 MHz IBW），高性能的三階段CFR發動機，集成的數字下調和數字上轉換都能夠支持多達八個組件載體。該設備適用于帶有小型單頻段，多波段，TDD Massive Mimo和TDD/FDD的此類部署/應用程序。圖9顯示了一個高級功能框圖。

圖9。ADRV9040高級功能框圖。

基于ZIF的體系結構

ADRV9040的發射和接收信號路徑使用零IF（ZIF）體系結構，該體系結構提供寬帶寬度，具有適用于非連續多載波RU應用程序的動態范圍。 ZIF架構具有低功率加上RF頻率和帶寬敏捷性的好處。該體系結構提供了比離散解決方案的尺寸，重量和功率優勢。該體系結構使OEM能夠設計最小，最輕的5G巨大MIMO收音機40％輕40％，能源效率高約10％。對完整的小信號無線電板的分析表明，ZIF體系結構可以在RF BOM與簡單導數RU上節省大量成本（PER 32T32R）。

零IF體系結構還以LO頻率傳輸能量。由于智商混合和數據路徑的差異，引入了正交和LO泄漏誤差（例如，載體不集中在LO上）（例如，兩個混合器從未真正具有相同的特征）。在多載波和不對稱載體應用中，這是一個更大的問題。為了減少這種不需要的發射，收發器具有TX LO泄漏校正算法，該算法都用于初始校準，然后是在運行時操作期間使用的跟蹤校準。

CFR塊

該設備的CFR有助于保持PAS線性。這種低功率CFR發動機可幫助設計人員降低輸入信號的峰值與平均值，從而使效率更高的傳輸線路UPS。如前所述，校正峰的光譜再生始終是CFR的關注點。至關重要的是，ADRV9040在優化算法中起著至關重要的作用，以確保CFR塊的影響與OEM的系統規格保持一致。理想的CFR塊具有非常低的延遲，零缺少峰值。

圖10顯示了5G新無線電（NR）信號上PAPR降低的結果。 PER-CFR（左）圖顯示了峰值壓縮，這是由輸出信號的CCDF（黃色痕跡）以陡峭的速率掉落的，而隨著PAPR的增加，高斯參考ccdf（綠色痕跡）的速度較高。另一方面，CFR后（右）圖顯示了其CCDF與高斯信號相似的5G NR信號。

圖10。5GNR信號在應用CFR之前和之后。

通過從檢測到的峰中減去預定的脈沖以使信號在PA的線性范圍內，使用脈沖取消技術的變體實現此CFR。 CFR塊由CFR發動機的三個副本組成，每個副本都使用檢測閾值來檢測峰和校正閾值，而檢測到的峰被減弱。這些頻譜校正脈沖從數據流中減去，以使信號在PA的線性范圍內。校正脈沖需要在光譜上形狀，以將噪聲泄漏到相鄰的頻段中。 ADRV9040可以容納兩個校正脈沖，同時對應于設備上的兩個不同的載波配置。這些校正脈沖可以預加載，并允許設備在即時兩種載波配置之間進行切換。

DPD

該設備包含一個完全集成的低功率DP??D引擎，用于RF信號鏈線性化應用。該引擎提供行業領先的DPD性能。如所討論的，介紹記憶的最通用方法是使用Volterra系列。該DPD發動機基于廣義記憶多項式（GMP）和動態偏差（DDR）的縮寫實現，它們是眾所周知的Volterra系列的廣義子集。在收發器的用戶指南和其他設計側支中詳細討論了ADRV9040中使用的廣義內存多項式。通過DPD執行器硬件將逆PA模型（PA ^-1^ ）應用于插值數字基帶樣本。專用的嵌入式^ARM?Cortex?^^^A55處理器用于計算GMP系數。 DPD執行器是可編程的多項式計算器。圖11顯示了應用于插值數字基帶樣品的PA ^-1模型。^

圖11。將逆PA模型應用于插值的數字基帶樣品。

該DPD算法支持間接學習和直接學習DPD機制，用于提取DPD模型系數。間接學習涉及使用觀察接收器數據作為預測與參考相對應的輸入樣本的參考，而直接學習涉及使用DPD前執行器發送信號作為參考以最大程度地減少觀察到的參考數據之間的誤差。兩者之間的區別在于，間接學習算法是時間效率的，但是直接學習算法更準確，因為它需要更長的時間來收斂。在不需要DPD的系統應用中，ADRV9040提供了一種通過GPIO控制繞過預性的機制。圖12顯示了在20 MHz LTE信號中使用DPD后ACLR的功率譜密度提高。通過在右圖上使用DPD，消除了左圖所示的ACLR的障礙。

圖12。功率光譜密度顯示DPD后ACLR的改善。

電源管理注意事項

使用正確的功率解決方案設計ADRV9040對于避免在TDD接收到傳輸過渡和實現最佳RF性能的問題之類的問題至關重要。 ADI SilentSwitcher?^技術^提供了多種區別，包括高切換頻率，超低RMS噪聲和點噪聲。無聲切換器3功率設計需要更少的組件，小的PCB足跡（尺寸），最重要的是一個更快且控制良好的瞬態沉降時間，從而導致超值EMI排放。圖13顯示了帶有一些建議的功率IC的宏基站的高級框圖，LT8627SP 和ADM7172 用于為ADRV9040電壓導軌提供動力。

圖13。具有ADRV9040功率解決方案的宏RRH的系統級框圖。

ADS10-V1EBZ和ADRV904X-MB/PCBZ評估平臺

ADRV9040評估平臺有助于建立一個簡單明了的框架，用于測試用戶的設計。圖14呈現 ADS10-V1EBZ （主板）和ADRV904X-MB/PCBZ評估板。請聯系您的ADI銷售代表以訂購您的評估系統。 RF設計師只需要將其設備連接到Eval平臺即可捕獲不同的圖，而ADRV9040進行了繁重的舉重，以通過Data Byte將其寫入其寄存器中找到最佳性能配置。

圖14。ADS10-V1EBZ（主板）和ADRV904X-MB/PCBZ評估板。

結論

電信技術的進步能夠通過載體聚合來提高數據速率通信和提高光譜效率，這也有助于PAPR的增加。但是，通過將CFR和DPD功能集成到專業設計的ADRV9040收發器中，無線電設計過程得到了簡化，從而減少了RF材料清單（BOM）成本，董事會尺寸，重量，重量和功耗，而不是常規的FPGA實施。隨著大量無線基站和全球部署的遠程單元，提高的功率放大器效率可以顯著降低服務提供商的能源和冷卻成本。這不僅加速了上市時間，還可以幫助降低運營費用，并確保部署在網絡中時的合規性。

作者

Hamed M. Sanogo

Hamed M. Sanogo是ADI公司全球應用部門的云和通信終端市場專家。Hamed擁有密歇根大學迪爾本分校的電子工程碩士學位，之后還獲得了達拉斯大學的工商管理碩士學位。在加入ADI公司之前，畢業后的Hamed曾在通用汽車擔任高級設計工程師，并在摩托羅拉系統擔任過高級電氣工程師以及Node-B和RRH基帶卡設計師。在過去的17年里，Hamed擔任過不同的職務，包括FAE/FAE經理、產品線經理，目前是通信和云終端市場專家。

附英文原文給大家參考：

Simplifying Your 5G Base Transceiver Station Transmitter Line-Up, Design, and Evaluation

摘要

With wireless communication standards such as LTE and 5G, the emphasis on higher data rates and spectral efficiency has driven the wireless original equipment manufacturers (OEMs) to adopt new transmission formats such as orthogonal frequency division multiplexing (OFDM). However, these signals, with large fluctuations in their envelopes, are especially vulnerable to nonlinear power amplifier (PA) distortions due to their high peak-to-average power ratios (PAPR). With this high PAPR signal, a PA nonlinearity can produce substantial signal distortions that increased bit error rates (BERs) and decreased signal-to-noise ratio as a result. This article reviews PAPRs, where they originate, how they can break down the RF components of the transmit line-up, and how to get rid of them or at least mitigate their effects on the signal chain.

Introduction

The newer modulation formats, such as OFDM, and various forms of quadrature amplitude modulation (QAM) have large fluctuations in their signal envelopes. This creates a high PAPR in the signal. Playing a high PAPR signal on a nonlinear PA generates spectral regrowth. Spectral regrowth refers to new frequencies that are caused by gain compression and were not in the original input. The high PAPR causes in-band distortion, which degrades the BER performance of the entire system. We will discuss a solution to help find the right system trade-off between efficiency and linearity using digital predistortion (DPD) and crest factor reduction (CFR) engines.

OFDM Modulation—Everyone Is Doing It!

In LTE and 5G systems, carrier aggregation, which is transmitting several carriers in parallel, is used to increase bandwidth and data rate. These networks leverage OFDM modulation, a very proficient and widely used multicarrier transmission technique that enables better spectral efficiency and reduces the impact of multipath reflections on the receiver’s ability to demodulate the signal. With OFDM, the final waveform is an orthogonal summation of subcarriers that carry information, where each subcarrier has its own center frequency and modulation scheme. In the time domain, sometimes the peaks of these subcarriers can align to produce an aggregate large OFDM waveform peak. A unique feature of OFDM is that the subcarrier waveforms are orthogonally combined such that the null (or zero amplitude) of one subcarrier coincides with the peak of other subcarriers as shown in Figure 1. This provides a relatively efficient use of the channel bandwidth, resulting in improved spectral efficiency compared to traditional single-carrier modulation.

Figure 1. Multicarriers OFDM subcarriers waveforms.

OFDM has several other benefits including its robustness against multipath fading. However, one of the major problems with OFDM modulation is that the transmitting waveforms suffer from a high PAPR. Figure 2 shows the PAPR of various common mobile technologies or modulation types. One can observe that the typical PAPR has been steadily increasing as new standards or modulation technologies have emerged.

Figure 2. Typical PAPR for various modulation technologies.

PAPR in OFDM Signals

As noted, carrier aggregation, enabled by OFDM modulation, is used to increase the bandwidth and data rate in 5G systems. OFDM also results in a signal whose envelope is nonconstant, and this can lead to high PAPR, which can contribute to system damage. If the RF power components in the RF signal line-up, especially the PA, are not suitably specified to handle the expected voltage peaks, these components can fail. A large PAPR reduces the efficiency of the PA by driving it deep into saturation, its nonlinear operating region, leading to distortion that results in spectral spreading of the signal. The linearity of the PA has always been a critical design issue for nonconstant-envelope digital modulation schemes. Figure 3 shows a time domain LTE 64-QAM signal captured at the ADRV9040’s transmitter output.

Figure 3. An illustration of orthogonal summation of subcarriers causing large peaks.

Complementary Cumulative Distribution Function

Due to its form, an OFDM signal requires a statistical approach for proper measurement. The complementary cumulative distribution function (CCDF) is used to evaluate the PAPR reduction performance in an RF signal chain. Figure 4a shows the transmitted waveform of an LTE downlink 10 MHz bandwidth and 64 QAM subcarrier modulation signal. The CCDF in Figure 4b shows that the signal power exceeds the average by at least 7.4 dB for 0.01% of the time. The theoretical maximum peak occurs at 0% probability, which is undefined on this plot. The trace intersects the x-axis (0.01%, or a probability of 10e ^-4^ ) at a PAPR of about 7.4 dB. This would indicate that one sample out of every 10,000 would be expected to exceed the average power by more than 7.4 dB.

Figure 4. CCDF of an LTE downlink with 10 MHz bandwidth and 64 QAM subcarrier modulation.

Upon taking a closer look at the CCDF graph, observe that the y-axis is cumulative probability and is usually plotted on a log scale; and the x-axis is power plotted in dB. The graph displays the probability or the percentage of time that a signal power is at or above the average power. Essentially, for each power level, the CCDF plot depicts the amount of time the signal spends above the average power level. As the CCDF curve moves to the right, the ratio of our peak power to the average power increases.

The CCDF plot verifies linear operation and is more often measured immediately after a PA. It can give a more accurate depiction of signal compression compared to the commonly used method of tracking changes in gain at differing power levels. The statistical analysis of crest factor occurrence makes it a valuable tool for designers to assess the impact of amplifier compression on the system’s BER and/ or error vector magnitude (EVM).

So, Why Is PAPR Important?

PAs are nonlinear in nature and exhibit a trade-off between linearity and efficiency. The common nonlinear problems are gain compression and phase distortion, including in-band and out-of-band distortions. Each of these factors degrades the BER performance of the system, as well as creates out-of-band spectral regrowth, which leads to adjacent channel interference, and violates out-of-band emissions standards mandated by regulatory bodies.

While testing a PA, the input amplitude is gradually increased until the measured ratio decreases by 1 dB, representing 1 dB gain compression. The 1 dB compression point is a key figure of merit that provides RF designers with assumptions about their amplifier’s performance. Essentially, the 1 dB compression point of an amplifier is defined as the output power at which the device’s gain drops by 1 dB from its small-signal value. This parameter is commonly used as a reference point for the beginning of amplifier nonlinearity and is approximately equal to the maximum useable peak output power for the amplifier. This is why many RF designers typically estimate their PA’s maximum operating output power to be a few dB lower than its 1 dB compression point. This makes finding the 1 dB compression point of a PA a crucial exercise, in order that a signal with a high PAPR is never allowed to saturate the PA. Another name for PAPR is crest factor. Figure 5 shows the AM-AM curve with the 1 dB compression point shown.

Figure 5. The AM-AM curve with the 1 dB compression point.

Now that the designer has evaluated the PA and identified its 1 dB compression point, they need to operate the PA in its linear region with an input power back-off (for example, operate the PA at a lower power within the linear portion of its operating curve) to avoid the spectral growth, right? Well, not exactly!

Simply backing off the input, far from the PA’s saturation point, can certainly help avoid all the nonlinear problems discussed, but results in very low efficiencies and increases heat dissipation. Solving this low efficiency problem by increasing the power consumption of the system is not a viable trade-off. As seen in Figure 2, as the standard bodies got innovative with new modulation schemes to make better use of the existing spectrum, this has resulted in signals with higher and higher levels of crest factor. So, using a PA back off implementation strategy would not work in the long run. The next sections of this article will discuss two implementation strategies that, when combined, will operate the PA up to its saturation point while still maintaining a good linearity and significantly increasing its efficiency. The first uses an amplitude clipping technique for PAPR reduction and the second method is to linearize the nonlinear response of a PA over its intended power range.

The Two Features of a Highly Successful Digital Front-End Solution

In a wireless digital front-end (DFE) system, a broad range of subsystems are covered, including DPD, digital upconversion (DUC), digital down-conversion (DDC), and CFR. There are also other important areas such as DC-offset calibration, pulse-shaping, image rejection, digital mixing, delay/gain/imbalance compensation, error correction, and other relevant blocks. The data captured at the output of the PA is utilized by the DPD circuit to linearize the PA output. DPD improves system linearity by allowing PAs to operate more efficiently, while CFR helps limit the signal PAPR. The DPD engine is used after using CFR to reduce the dynamic range of the signal and allows the PA to be operated above the linear region. While each of these blocks covers key features of the DFE, this section of the article will only focus on the CFR and DPD blocks.

Crest Factor Reduction

Most of the input signal of an OFDM waveform will be within the linear range of the PA. However, as previously shown, the signal has peaks that may exceed the PA’s linear operating range, which invites long-term reliability concerns due to their contribution to system damage. Again, it is highly desirable to drive the PA at the highest input power possible without having it saturate. To avoid saturation due to the peaks, CFR is used, where instead of attenuating the whole signal, only the portions of the signal that are above the PA’s linear range are attenuated. In short, CFR assists in keeping the PA linear.

When peaks are suppressed, this results in a constant output power, thus ensuring that the signal remains within the PA’s linear range. Note that CFR is not a linearization technique, but it is rather an efficiency improvement scheme. With its most effective implementation, CFR eliminates the peaks of the transmit signal to reduce the peak-to-average ratio while complying with the desired spectral emission mask, adjacent channel power ratio, and EVM specifications. Figure 6 shows detected signal peaks above a threshold level. The magnitude of the peaks is reduced to below some target value. This is usually followed by filtering to reshape the signal spectrum.

Figure 6. Detected signal peaks above the threshold level are reduced.

A downside of CFR is that clipping leads to in-band signal distortion, resulting in BER performance degradation, and out-of-band radiation, which imposes out-of-band interference signals to adjacent channels. In short, the consequence of clipping is poor signal ACLR and EVM. Filtering the clipped signal is often used to reduce out-of-band radiation at the cost of peak regrowth.

Digital Predistortion

With DPD, the PA can be pushed to operate up to the saturation region, without compromising its linear characteristics. DPD allows RF designers to operate their systems in the efficient, yet nonlinear, region of a PA while retaining the transmit signal linearity required of the OFDM modulation. In other words, with DPD the PA’s linear region is extended. The DPD engine produces predistorter coefficients by modeling the inverse AM-AM and AM-PM characteristics of the PA. Essentially, DPD focuses on improving the quality of a signal that the PA produces when operating at its peak efficiency point. DPD aims to introduce inverse nonlinearities that compensate for the PA gain. This is a technique for improving the linearity of a nonlinear PA by introducing precise antidistortion into the input waveform that compensates for the PA’s in-band nonlinear products. Figure 7 shows the concept of DPD for linearizing a PA response.

Figure 7. Generic concept of DPD for linearizing the PA response: (a) Typical AM-AM curve showing the overall linear region is in green; (b) basic concept of DPD and how it improves power amplifier efficiency.

It works on the principle of predistorting the transmitted data in the digital domain to cancel the distortion caused by PA compression in the analog domain. The approach to DPD can range from simple solutions such as a basic lookup table (LUT) to a more complex real-time signal processing approach. DPD implementations can be classified into memoryless models and models with memory.

Memoryless DPD

Memoryless DPD corrects the amplitude and phase of the IQ samples based on the current sample only. PAs that are strictly memoryless can be characterized by their AM-AM and AM-PM conversions. This instantaneous nonlinearity is usually characterized by the AM-AM and AM-PM responses of the PA, where the output signal amplitude and phase deviation of the PA output are given as functions of the amplitude of its current input. Therefore, a memoryless PA can be characterized by its AM-AM and AM-PM responses. These measurements are used to create LUT data that relates every input power/phase combination to the power/phase required to produce the desired linear output. The advantage of memoryless DPD is that it can be implemented relatively straightforward as a lookup table. Figures 7a and 7b show AM-AM response of a PA with and without DPD correction applied to a 2× 100 MHz, 400 MHz bandwidth 4096 samples dataset.

Figure 8. AM-AM response of a PA with and without DPD on a 2× 100 MHz, 400 MHz bandwidth signal.

DPD with Memory

As the transmit signal bandwidth gets wider, PAs will begin to exhibit memory effects. These are nonuniform frequency responses in certain components like the biasing network, decoupling capacitors, power supply circuitry, or can be attributed to thermal constants of the active devices. As a result, the current output of the PA then depends not only on the current input but also on past input values. When this is the case, this means that the PA has become a nonlinear system with memory. DPD with memory corrects amplitude and phase of IQ data samples based on several previous samples and their interdependencies. The response of the PA generally does not only depend on the current signal amplitude, but also on the amplitudes of the previous samples. Thus, the digital predistorter would also need to have memory structures—it is the mathematical backbone of DPD. The Volterra series is the most general polynomial type of nonlinearity with memory and used to model nonlinear systems with memory. Therefore, the most general way to introduce memory is to use the Volterra series. For an elaborate discussion on the details of the math behind modeling PA distortion with the Volterra series, which is beyond the scope of this article, please refer to Masterson (2022).^2^

Framework for Designing 5G RF Signal Chains with Ease

The ADRV9040 RF transceiver provides a streamlined framework for designing, implementing, and testing the RF signal chain lineup of a 5G communication system with ease. A discrete massive MIMO system requires four chip levels in its discrete form of deployment, including an RF transceiver, DFE FPGAs, baseband FPGA/ASIC, and a control FPGA. Since this transceiver comes with the DFE integrated, it eliminates the need for several FPGAs used in competitive discrete solutions where the DPD, CFR, DUC, and DDC blocks are implemented in computer code. The FPGA implementation is typically costly and power hungry. This highly integrated RF transceiver helps eliminate such power-hungry dedicated FPGAs. In this section of the article, we highlight this RF transceiver and a proposed framework for checking a typical PA gain line-up and doing a sanity check for noise limits by implementing register writes inside the device.

This highly integrated, radio frequency agile transceiver system on chip (SoC) offers eight transmitters, two observation receivers for monitoring transmitter channels, eight receivers, integrated local oscillator (LO) and clock synthesizers, and digital signal processing functions to provide a complete transceiver with significant digital front-end capability. The device provides the high radio performance and low power consumption demanded by cellular infrastructure applications such as small cell radio units (RUs), macro 4G/5G RUs, and massive MIMO RUs. The complete transceiver subsystem includes automatic and manual attenuation control, DC offset correction, quadrature error correction, and digital filtering. The transceiver has a fully integrated digital front-end that supports a few key blocks, including DPD (up to 400 MHz IBW), a high performance three-stage CFR engine, integrated digital downconversion and digital upconversion both capable of supporting up to eight component carriers. The device is suitable for such deployments/applications with small cell single-band, multi-band, TDD massive-MIMO, and TDD/FDD in macro-RU equipment. Figure 9 shows a high level functional block diagram.

Figure 9. An ADRV9040 high level functional block diagram.

ZIF-Based Architecture

The transmit and receive signal paths of the ADRV9040 use a zero-IF (ZIF) architecture that provides a wide bandwidth with dynamic range suitable for noncontiguous multicarrier RU applications. The ZIF architecture has the benefits of low power plus RF frequency and bandwidth agility. This architecture provides size, weight, and power advantages over discrete solutions. The architecture enables OEMs to design the smallest and lightest 5G massive-MIMO radios 40% lighter and ~10% more energy efficient. Analysis of a complete small signal radio board shows that the ZIF architecture enables significant cost savings (per 32T32R) on RF BOM vs. simple derivative RUs.

The zero-IF architecture also transmits energy at the LO frequency. The quadrature and LO leakage errors (for example, carriers are not centered on LO) are introduced due to differences in IQ mixing and data paths (for example, two mixers never really have the same characteristics). This is an even bigger issue in multicarrier and asymmetric-carrier applications. To reduce this undesired emission, the transceiver has a Tx LO leakage correction algorithm, which is used both for initial calibration followed by a tracking calibration used during runtime operation.

CFR Block

The device’s CFR assists in keeping PAs linear. This low power CFR engine helps designers reduce the peak-to-average ratio of the input signal, enabling higher efficiency transmit line ups. As mentioned previously, spectral regrowth of corrected peaks is always a concern with CFR. It is crucial to note that the ADRV9040 plays a vital role in optimizing the algorithm to ensure that the impact of the CFR block aligns with the system specifications of the OEM. The ideal CFR block has very low latency and zero missed peaks.

Figure 10 shows the results of a PAPR reduction on a 5G new radio (NR) signal. The pre-CFR (left) plot shows peak compression, which is indicated by the output signal’s CCDF (yellow trace) falling off at a steeper rate than the input, Gaussian reference, CCDF (green trace) as the PAPR increases. On the other hand, the post-CFR (right) plot shows a much improved 5G NR signal where its CCDF is similar to that of the Gaussian signal.

Figure 10. 5G NR signal before and after applying CFR.

This CFR is implemented using a variation of a pulse cancellation technique by subtracting a precomputed pulse from the detected peaks to bring the signal within the PA’s linear range. The CFR block consists of three copies of CFR engines, each of which uses a detection threshold to detect the peaks and a correction threshold to which the detected peaks are attenuated. These spectrally shaped correction pulses are subtracted from the data stream to bring the signal within the PA’s linear range. The correction pulse needs to be spectrally shaped to manage the noise leakage into adjacent bands. The ADRV9040 can hold two correction pulses corresponding to two different carrier configurations on the device at the same time. These correction pulses can be preloaded and allow the device to switch between two carrier configurations on-the-fly.

DPD Block

The device includes a fully integrated, low power DPD engine for use in RF signal chain linearization applications. This engine provides industry-leading DPD performance. As discussed, the most general way to introduce memory is to use the Volterra series. This DPD engine is based on an abbreviated implementation of generalized memory polynomial (GMP) and dynamic deviation reduction (DDR), which are generalized subsets of the well-known Volterra series. The generalized memory polynomials used in the ADRV9040 are discussed in detail in the transceiver’s user guide and other design collaterals. An inverse PA model (PA ^-1^ ) is applied to the interpolated digital baseband samples through the DPD actuator hardware. A dedicated embedded Arm^?^ Cortex^?^ A55 processor is used for the computation of the GMP coefficients. The DPD actuator is a programmable polynomial calculator. Figure 11 shows a PA^-1^ model applied to the interpolated digital baseband samples.

Figure 11. An inverse PA model is applied to the interpolated digital baseband samples.

This DPD algorithm supports both indirect learning and direct learning DPD mechanisms for extracting DPD model coefficients. Indirect learning involves using the observation receiver data as a reference for predicting the input samples corresponding to the reference, while direct learning involves using the pre-DPD actuator transmit signal as a reference to minimize the error between the observed and reference data. The difference between the two is that the indirect learning algorithm is time efficient, but the direct learning algorithm is more accurate since it requires a longer time to converge. In a system application where DPD is not needed, the ADRV9040 provides a mechanism to bypass predistortion through GPIO control. Figure 12 shows the power spectral density improvement in ACLR after the application of DPD for a 20 MHz LTE signal. The impairments that provoked the ACLR skirt shown on the left plot have been removed by using DPD on the right plot.

Figure 12. Power spectral density showing improvement in ACLR post-DPD.

Power Management Considerations

Designing the ADRV9040 with the right power solutions is paramount to avoiding issues such as the first symbol poor EVM (for example, Cyclic prefix) at the TDD receive-to-transmit transitions and to achieving optimum RF performance. The ADI Silent Switcher^?^ technology offers several differentiations, including high switching frequency, ultralow rms noise, and spot noise. A Silent Switcher 3 power design requires fewer components, a small PCB footprint (size), and most importantly a faster and well-controlled transient settling time, which results in ultralow EMI emissions. Figure 13 shows a high-level block diagram of a macro base station with a few suggested power ICs, LT8627SP and ADM7172 for powering the ADRV9040 voltage rails.

Figure 13. System-level block diagram of a macro RRH with the ADRV9040 power solutions.

ADS10-V1EBZ and ADRV904X-MB/PCBZ Evaluation Platform

The ADRV9040 evaluation platform facilitates the establishment of a simple and straightforward framework for testing the user’s design. Figure 14 presents the ADS10-V1EBZ (motherboard) and ADRV904X-MB/PCBZ evaluation board. Please contact your ADI sales representative to order your evaluation system. The RF designer only needs to connect his equipment to the eval platform to capture the different plots while the ADRV9040 does the heavy lifting to find the optimal performance configuration via data byte writes into its registers.

Figure 14. ADS10-V1EBZ (motherboard) and ADRV904X-MB/PCBZ evaluation board.

Conclusion

The advancements in telecom technologies that enable higher data rate communications and improved spectral efficiency through carrier aggregation also contribute to an increase in PAPR. However, by integrating CFR and DPD capabilities into the expertly designed ADRV9040 transceiver, the radio design process is simplified, resulting in reduced RF bill of materials (BOM) cost, board size, weight, and power consumption compared to conventional FPGA-based implementations. With a large number of wireless base stations and remote units deployed globally, improved power amplifier efficiency can significantly reduce energy and cooling costs for service providers. This not only accelerates time-to-market but also helps lower operational expenses (energy and truck rolls) and ensures compliance when deployed in networks.

關于作者

Hamed M. Sanogo

Hamed M. Sanogo是ADI公司全球應用部門的云和通信終端市場專家。Hamed擁有密歇根大學迪爾本分校的電子工程碩士學位，之后還獲得了達拉斯大學的工商管理碩士學位。在加入ADI公司之前，畢業后的Hamed曾在通用汽車擔任高級設計工程師，并在摩托羅拉系統擔任過高級電氣工程師以及Node-B和RRH基帶卡設計師。在過去的17年里，Hamed擔任過不同的職務，包括FAE/FAE經理、產品線經理，目前是通信和云終端市場專家。