【正文】 of the window assembly. The input and output ports are standard WR4 waveguide ( in in). Conical taper mode converters transition from the TE10 rectangular to TE11 circular waveguide. The windowconsists of a beryllium oxide (BeO) ceramic disk, metallized on the edge, brazed into a copper cylinder. The window assembly is clamped between the two mode converters, as shown in Fig. 1(a). The properties of the BeO disk are shown in Table I.Fig. 1. (a) Geometry of pillbox window. Taper mode converters transition between TE10 rectangular and TE11 circular waveguide. The circular waveguide length on one side of the window disk is L. (b) Optical micrograph of a brazed window assembly at 50 magnification. (c) Photograph of window assembly next to a . dime.A vector network analyzer is used to measure the plex reflection (S11) and transmission (S21) parameters of the windows. Testing in the range of 140–325 GHz is acplished by using two sets of mmW extenders, one in WR5 waveguide (140–220 GHz) and one in WR3 (220–325 GHz). Each window is assembled in a coldtest fixture that includes the mode converters and input/output WR4 waveguides, and taper transitions between WR5/WR3 and WR4 are connected to each test port. The total insertion loss of the waveguides and transitions is dB across the band and is not calibrated out of the measurement. The insertion loss of the ceramic disk itself is determined to be dB, which is below the noise threshold of our setup. Fig. 2 shows the S11 and S21 magnitudes for the windows chosen for the input and output ports of the TWT. Included for parison are 3D electromagnetic simulation results (HFSS) showing an optimized design for each window, in which the value of L is adjusted in the simulation to provide an ideal broadband match given the measured window thickness t. The windows were tuned toward this optimum using the procedure described in Section B below, achieving ?20 dB reflection over ～20 GHz bandwidth. The input and output window thicknesses are t = 292 177。 5μm, respectively. The pass band of the output window is ～10 GHz higher than that of the input window and somewhat wider. This window was chosen for the output in order to mitigate feedback oscillation from a highfrequency tail in the predicted gain curve of the amplifier .Fig. 2. S11 and S21 magnitude measurements for (a) input and (b) output windows. (Black crosses。 they are thus trapped in the vicinity of the window and evanesce into the waveguide on either side. Ghost mode resonances associated with perturbations such as offset or tilt misalignment of the window disk, nonuniformity of the window edge metallization/braze, or nonuniformity of the dielectric material can also be excited by waveguide propagating modes. Three ghost modes observed near 200–210 GHz and one near 230 GHz are TE41?, TE12?, TM11?, and TE51?like modes, respectively, and were intentionally moved outside our operating band in thedesign.An extended interaction klystron was used to test windows at W, 100% duty at 218 GHz for several minutes. No pulse distortion or reflection that would indicate breakdown was observed during the highpower testing.Fig. 3. Measured fractional power loss of (a) input window and (b) output window. The baseline 30%loss is due to the input/output waveguide transitions.Fig. 4. S11 measurements of output window during tuning procedure, between lapping operations.B. Window TuningThe window response was tuned by lapping the copper cylinder piece on each end face to adjust the circular waveguide length L, which was initially fabricated oversized to allow tuning. This was done prior to taking the final coldtest data shown in Fig. 2. Fig. 4 shows the S11 of the output window during the tuning procedure, measured between lapping operations. Each lapping removed ～10–30 μm of copper from one end of the cylinder, and lapping was alternated between the two ends to maintain symmetry. For this particular window, the tuning procedure widened the pass band by 15 GHz and improved reflection by 15 dB over the band in parison to the initially fabricated piece. The final L values for the two windows were in the range of 730–870 μm.III. CONCLUSIONThis brief has presented measurements of broadband pillbox vacuum windows that transmit mmW in the range of 210–235 GHz. Our coldtest data and simulations demonstrate systematic tuning of each window after fabrication to achieve broadband transmission close to an ideal design. We found it necessary to control the circular waveguide length to ≤ 10 μm tolerance. Another critical parameter was the variance in the as fabricated thickness of the BeO disks。通過傳輸和反射測試顯示在中心頻率處最少25GHz帶寬內(nèi)回波損耗優(yōu)于20dB。在218GHz出現(xiàn)100%功率傳輸。阻抗匹配條件使得輸出窗只能傳輸一定頻段的電磁波，在頻段外電磁波的大部分能量將被反射。常規(guī)矩形波導(dǎo)輸出窗是半波諧振窗，通常是在一段導(dǎo)波中填充一些介質(zhì)材料。這種結(jié)構(gòu)的響應(yīng)帶寬是窄帶的，因此不適應(yīng)要求傳輸帶寬 10%的寬頻放大器。在工作頻段，選擇合適的圓波導(dǎo)長度來消除進(jìn)出口錐形和窗片帶來的反射。隨著頻率的增加，達(dá)到寬帶阻抗匹配必要的所需制造誤差變得越來越困難。簡單來說，我們給出了在220GHz的寬帶錐形輸出窗的實驗測試結(jié)果。這種輸出窗是為行波管放大器在頻帶 范圍內(nèi)輸出功率超過50W而設(shè)計的。 圖1顯示這種窗的裝置示意圖。錐形模型轉(zhuǎn)換裝置將矩形波導(dǎo)的TE10模轉(zhuǎn)換為圓波導(dǎo)的TE11模。如圖1所示，這種輸出窗就是由這兩部分夾緊構(gòu)成。圖1 （a）錐形輸出窗的幾何模型。圓波導(dǎo)的長度為L。（c）放在硬幣旁的窗片組件照片用矢量網(wǎng)絡(luò)分析儀測量輸出窗復(fù)雜的反射系數(shù)（S11）和透射系數(shù)（S21）。兩種輸出窗都是在冷測裝置中將模式轉(zhuǎn)化器、輸入輸出端WR4矩形波導(dǎo)通過WR5/ WR3與WR4的錐形裝換連接到相應(yīng)測試端口。陶瓷窗片本身的插入損耗被確定為＜，這個結(jié)果低于我們所設(shè)置的噪聲閥值。通過包括三位電磁仿真結(jié)果（基于HFSS）的對比，在給定窗片厚度t的情況下通過調(diào)節(jié)圓波導(dǎo)的長度L可以得到最優(yōu)的設(shè)計模型。在輸入輸出端口窗片的厚度分別為t = 292 177。 5μm。將此輸出窗作為輸出端是為了減小在放大器預(yù)設(shè)增益線高頻尾端的反饋震蕩。部分損耗的峰值具有存在于輸出窗中諧振鬼模的特性，具有類似于圓波導(dǎo)的橫向諧振結(jié)構(gòu)，但是其頻率低于對應(yīng)空波導(dǎo)的截止頻率；因此他們被束縛在輸出窗附近并消失在波導(dǎo)兩側(cè)。在200~210GHz附近像TE41?, TE12?, TM11?,這種模式的三個鬼模和在230GHz附近的像TE51?這種模式的一個鬼模，在這種設(shè)計中是我們分別特意將其移出工作頻段的。（黑色十字線，黑色實線）HFSS的仿真結(jié)果表明基于測量窗片厚度而設(shè)計出的最優(yōu)輸出窗效果。其中最起碼有30%的損耗是由于輸入/輸出波導(dǎo)過渡帶引起的。如果沒有脈沖失真或反射，則表明高功率測試下故障將被觀察到。這樣做是為了將最后的冷腔測試數(shù)值表示在圖2中。每次將銅軸圓減小約10–30 μm厚，并且減小時保持兩端的對稱。最終確定兩個窗的L值都在730–870 。經(jīng)過冷腔測試和模擬優(yōu)化后，每個窗口實現(xiàn)在要求頻段內(nèi)的傳輸接近理想設(shè)計。另一個關(guān)鍵的參量為氧化鈹窗片的厚度，通過仿真可以預(yù)測到窗片厚度每改變20μm，整個窗的帶寬就相應(yīng)改變10GHZ。以上結(jié)果充分證明這是一種設(shè)計高頻毫米波電真空器件寬頻輸出窗的