JP6746051B2 - Electronic device housing structure - Google Patents

Description

The present invention relates to a housing structure for an electronic device capable of realizing communication between electronic devices housed inside by wireless communication.

Along with the development of cloud services, the number of data centers that store large amounts of data via networks is increasing. Many devices such as servers are used in data centers. Due to its nature, the data center needs to be operating 24 hours a day, 365 days a year, without any breaks. As a result, each data center consumes a large amount of power throughout the year. Therefore, reducing the power consumption of the entire data center is an urgent issue. The server itself consumes power and must be cooled to prevent the server from overheating and producing errors. Of the power consumed in the data center, not only the power consumed by devices such as servers but also the proportion of air-conditioning power for cooling the servers is very large.

The total global energy consumption of the data center was about 270 TWh in 2012, which is equivalent to about 2% of the global electricity consumption. It is predicted that this number will grow by about 4.3% each year. A major part of this energy consumption comes from the air conditioner, as mentioned above.

On the front side of a typical data center cabinet, a Category 6 Ethernet cable spanning a total of ten meters typically connects 40 servers to a central L2/L3 switch. It is said that

Such cabling essentially obstructs proper air flow and thus increases the cost of cooling energy.

On the other hand, 60 GHz band wireless communication, which can provide a wide bandwidth, has been recognized as a cost-effective alternative communication technology.

Moreover, replacing the cables wirelessly simplifies data center maintenance, including rerouting server connections whenever needed.

Therefore, recently, many research reports have reported the communication between the data center and the cabinet by using the radio of 60 GHz within the line-of-sight distance (for example, refer to Non-Patent Document 1).

However, there is little published prior work on intra-cabinet communication between the L2/L3 switch and each server.

In a general data center, a large number of server racks each accommodating a plurality of servers are densely installed in order to operate the servers efficiently. Typically, one server rack accommodates about 40 servers. Each server rack is arranged so that the cooling air from the cooling device flows.

JP 2006-217054 A

Y. Katayama, T. Yamane, Y. Kohda, K. Takano, D. Nakano, and N. Ohba, “MIMO Link Design Strategy for Wireless Data Center Applications,” IEEE WCNC, 2012.

However, the operation of the server in such a data center has the following problems. In the server rack, it is necessary to connect each server and the switch for communication. This connection is made by a communication cable such as an Ethernet (registered trademark) cable. However, it is said that if there are many servers, these cables will be complicated and maintenance of the server will be complicated, and the cooling air flow will be obstructed by the cables, and the cooling efficiency will be significantly reduced.

One way to solve these problems is to make the communication between the server and the switch in the data server wireless.

Since the server rack is usually made of metal, it is relatively easy to trap radio waves. However, when wireless communication is performed in a space confined by metal, radio waves are reflected multiple times on a metal surface such as a wall surface in a server rack, which is an extremely severe environment for wireless communication. In the presence of multiple reflections, many delayed waves are received in addition to the main wave. As a result, interference due to a delayed wave occurs between modulation symbols arranged in time series.

OFDM (Orthogonal Frequency Division Multiplexing) is a modulation method that is strong against interference between symbols due to delayed waves. In OFDM modulation, a cyclic prefix is added to the beginning of each symbol to provide a guard interval period. As a result, it is possible to avoid the interference of delayed waves that arrive within this period. In particular, as one of the influential communication methods among the communication methods in the server rack, there is OFDM modulation of the IEEE 802.11ad standard in the 60 GHz band. In this OFDM modulation, the guard interval period is 48.4 ns. Therefore, if the delay wave of a large level is delayed within 48.4 ns with respect to the main wave, communication can be performed by ignoring intersymbol interference. However, IEEE 802.11ad is a standard that is intended for indoor use, and is not intended for use in a narrow metal space such as a server rack.

As a measure to solve this problem, an electromagnetic wave absorber is provided on the inner surface of the server rack.

However, it is extremely difficult in practice to arrange the absorber so as to cover the entire metal surface, and the delayed wave still remains because the radio wave is reflected by the metal surface that cannot be covered with the absorber. Therefore, it is necessary to effectively arrange the absorber within a limited range to reduce the delayed wave as much as possible and improve the transmission characteristics. However, in the server rack, the radio wave propagation environment inside the server rack changes due to the position of the stored server, the position of the antenna, the position of cables and other objects, and the expansion and contraction of objects due to temperature changes. In particular, at a high frequency, there is a large change, and it is difficult to effectively arrange the absorber in a fixed manner at the time of manufacturing the housing.

An object of the present invention is to provide an electronic device capable of adaptively changing the arrangement of absorbers on the inner surface of the housing when a plurality of electronic devices are housed in a metal housing such as a server rack. It is to provide a containing structure.

According to one aspect of the present invention, there is provided a housing structure for an electronic device, which has a housing defining a space for housing a plurality of electronic devices therein, and includes a front half part having an opening and a closed rear half part. The housing has a housing, and the inner space of the housing is divided into a front half and a rear half by accommodating a plurality of electronic devices therein, and the rear of the plurality of electronic devices is a rear half. Each of the plurality of electronic devices has a communication module for transmitting and receiving a radio signal of a predetermined frequency on the back side and is provided in each of a plurality of regions provided on the inner surface of the housing. comprising a are active region, each active region has an absorbent member provided on the back side of the installed that variable material constant member and a variable material stereotactic number member to direct the front inside the housing, each plurality The surface of the variable material constant member on the inside of the housing can switch its function as a transmissive body or a reflective body with respect to a radio signal of a predetermined frequency under the control of the control section. During the calibration period, the state of the surface of the plurality of variable material constant members is adaptively switched based on the communication metric when switching to the transmissive body or the reflective body, and the communication metric is the communication communication of the plurality of electronic devices. The median value of the capacity, and the control unit sets the surface states of the plurality of variable material constant members so that the median value of the communication capacity is maximized.

According to still another aspect of the present invention, there is provided a housing structure for an electronic device having a housing defining a space for housing a plurality of electronic devices therein, the first half having an opening, and the second half closed. The housing has a housing, and the inner space of the housing is divided into a front half and a rear half by housing the plurality of electronic devices therein, and the plurality of electronic devices have a rear surface. Arranged so as to face the second half side, each of the plurality of electronic devices has a communication module for transmitting and receiving a radio signal of a predetermined frequency on the back side, and each of the plurality of electronic devices is provided in a plurality of regions provided on the inner surface of the housing. An active region is provided, and each active region has a variable material constant member installed so that its front surface faces the inner surface side of the housing and an absorbing member provided on the back surface side of the variable material localization member. The surfaces of the plurality of variable material constant members on the inner side of the housing can switch their functions as a transmissive body or a reflective body with respect to a radio signal of a predetermined frequency under the control of the control section. Is adaptively switched based on the communication metric when the surface state of the plurality of variable material constant members is switched to the transmissive body or the reflective body during the calibration period. It is the median value of the condition number, and the control unit sets the surface states of the plurality of variable material constant members so that the median value of the condition number is minimum .

Preferably, for each of the plurality of variable material constant members, a plurality of sets of states, each of which is a transmitter or a reflector, is predetermined, and the control unit calculates a communication metric for each of the plurality of sets of states. , The state of the surface of the plurality of variable material constant members is set to the optimum state of the set of the plurality of states.

Preferably, the predetermined frequency is a millimeter wave band, and the electronic device is a server.

Preferably, the housing has a rectangular parallelepiped shape including a top plate, a bottom plate, two side plates, a back plate, and a front door provided with an opening, and the plurality of servers are arranged in an internal space defined by the housing. , Is housed and fixed in the housing so as to be separated from the back plate by a certain distance, whereby the back half plate is defined by the back plate, the bottom plate and the top plate, the two side plates, and the back faces of the plurality of servers. The variable material constant members are installed on the inner surfaces of the two side plates, respectively.

According to the present invention, it is possible to efficiently suppress the influence of delayed waves in wireless communication in a housing that needs to ensure a passage for cooling air.

Further, according to the present invention, it is possible to adaptively change the arrangement of the absorbers on the inner surface of the housing and effectively suppress the influence of the delayed wave.

It is a conceptual diagram for demonstrating the state where the server was accommodated in the server rack and the inside. It is a figure for explaining the state where a plurality of servers were stored in a server rack, and the external appearance of each server. 3 is a functional block diagram for explaining a transmission configuration of a communication module 94. FIG. It is a functional block diagram for explaining a receiving configuration of a communication module 94. It is a conceptual diagram for demonstrating the size of a server rack, and the antenna direction of an access point (AP). FIG. 6 illustrates an active area located within a cabinet of a server rack backplane space. FIG. 6 is a diagram for explaining a configuration of a metamaterial that is a variable material constant member described in known document 1. It is a figure which shows the simulation result of a wireless ray tracing. 3 is a plot of corresponding wireless ray tracing software. It is a top view for explaining arrangement of an Rx antenna and a Tx antenna used for simulation. It is a figure which shows a simulation parameter. FIG. 11 is a diagram showing a distribution of communication capacities when antennas are arranged as shown in FIG. 10. It is a figure which shows the cumulative distribution function (CDF) of the communication capacity with respect to each of a set of active areas. It is a figure which shows the simulation result of CDF with respect to the condition number about each of the set of active regions. It is a flowchart which shows the 1st procedure of searching the optimal activation group of an active region about a certain server rack among the control procedures of a control part. It is a flowchart which shows the 2nd procedure of searching the optimal activation group of an active region about a certain server rack among the control procedures of a control part.

Hereinafter, a housing structure for an electronic device according to an embodiment of the present invention will be described with reference to the drawings. In the following embodiments, components and processing steps given the same reference numerals are the same or equivalent, and the description thereof will not be repeated unless necessary.

As described below, for the housing structure of the electronic device of the present embodiment, the case where the electronic device is a server and the housing structure is a server rack will be described as an example, and the OFDM system is used as the communication system. Shall be adopted. The communication frequency is assumed to be in the millimeter wave band, and transmission at 60 GHz is considered as an example.

FIG. 1 is a conceptual diagram for explaining a server rack and a state where a server is housed inside the server rack.

As shown in FIG. 1A, a typical server rack 30 has a rectangular parallelepiped shape with a height of about 2.2 m, a width of about 0.7 m, and a depth of about 1.2 m. A large number of thin servers are housed inside this. The entire server rack 30 includes a metal bottom plate 40, a top plate 42, two side plates 44 and side plates 46, a back plate 48, and a front door 50. The front door is provided with an opening 52, and in many cases, the opening 52 has a mesh shape. The cooling airflow is supplied into the server rack 30 through the opening 52.

With reference to FIG. 1B, when a large number of servers are housed inside the server rack, a substantially closed space 54 is formed between the server back surface and the inner surface of the back surface of the server rack. When performing wireless communication in the server rack, it is considered that the radio waves are trapped in this space and do not leak to the outside.

FIG. 2 is a diagram for explaining a state in which a plurality of servers are housed in a server rack and the appearance of each server.

FIG. 2A shows the inside of the server rack according to the present embodiment, together with the inner surface of the server rack and the vicinity of the upper and lower centers of the multiple servers 60 housed inside the server rack. The L2/L3 switch (access point) 62 connected to each server is installed near the upper and lower center of the server rack, and the servers 60 are installed above and below the server 60 without any gap. The server 60 is housed so that its rear surface faces the rear surface side of the server rack and a certain distance remains between the inner surface of the rear surface of the housing. With this arrangement, a space (second half) is formed between the rear surface of the server and the inner surface of the rear portion of the server rack, and wireless communication is performed inside this space. The housing of the server rack 30 is made of metal, and a mesh-shaped front door is provided on the front surface thereof so that the cooling airflow can flow.

In this embodiment, the size of the holes in the mesh of the front door is set to less than half a wavelength so that the leakage of radio waves of the used frequency is minimized. For example, when the frequency used is 60 GHz, the size of the holes in the mesh is less than 2.5 mm.

By thus accommodating the server and the L2/L3 switch 62 inside the server rack, the internal space of the server rack is separated into a front half on the opening side and a closed rear half on the back side. The rear half space is defined by the bottom plate, the top plate, the two side plates, and the back plate of the server rack.

As will be described later, mounting terminals for a communication module that performs wireless communication are provided on the back surfaces of the server 60 and the L2/L3 switch 62.

An inner surface of the bottom plate 40 and the top plate 42 (a surface facing the inside of the server rack), which is a region between the back surface of the server and the inner surface of the rear portion of the server rack is referred to as a region 80, and the top plate portion is referred to as a region 82. I will decide. Regarding the bottom plate part, there is a case where the bottom plate of the server rack 30 is not provided and the concrete floor is visible. In the following simulations, description will be given assuming that such a configuration does not have a bottom plate.

Referring to FIG. 2B, the communication module 94 has a USB interface. A USB terminal 92 is provided on the back surface 90 of the server 60, and a communication module 94 is attached to it. A similar USB terminal is also provided on the back surface of the L2/L3 switch 62. The communication module 94 is attached to each USB terminal of the server 60, and the communication module 94 is also attached to the USB terminal of the L2/L3 switch 62.

FIG. 3 is a functional block diagram for explaining the transmission configuration of the communication module 94.

On the transmit side, scrambler 102 receives the transmitted data signal and scrambles the information bit stream of the data portion to reduce the occurrence of long sequences of 1s or 0s. The FEC unit 104 encodes the input information bits to generate a corresponding encoded stream. The FEC unit 104 is, for example, a BCC (binary convolutional code) encoder.

The puncturing unit 106 generates a code having a desired coding rate. The generated codeword is distributed to each stream by the parser 108. Each stream is interleaved by each interleaver 110.1-110.4, and mapping to each constellation point is performed by each mapper 112.1-112.4. The signals from the mappers 112.1 to 112.4 are subjected to transmission MIMO processing by integrating the weighting factors by the transmission beam forming unit 114, and then inverse Fourier transformed by the IFFT units 116.1 to 116.4. GI (guard interval) is added, OFDM modulation is performed, and the signals are transmitted from antennas 120.1 to 120.4 via an analog front end (not shown).

FIG. 4 is a functional block diagram for explaining the reception configuration of the communication module 94.

On the reception side, the reception signals of the respective antennas are received via the antennas 202.1 to 202.4 and an analog front end (not shown), and the timing coarse adjustment units 204.1 to 204.4 and the timing fine adjustment units 206.1 to 206. By 4, the deviation is removed between the frequency on the transmitter side and the frequency on the receiver side.

The FFT units 208.1 to 208.4 remove GI and perform OFDM demodulation. The CSI evaluation units 210.1 to 210.4 generate CSI including the reception state based on the channel state information (CSI) from the access point. The CSI is returned to the access point as CSI feedback. At the access point, the transfer function is derived based on it, thereby performing more accurate beamforming.

The beam forming unit 212 executes a beam forming process by a ZF (Zero-Forcing) algorithm or the like on the receiving side to perform a receiving MIMO process. The signal after the reception MIMO process is demodulated by demappers 214.1 to 214.4, deinterleaved by deinterleavers 216.1 to 216.4, merged into one stream by deparser 218, and depunctured section 220, The error is corrected by being sent to the Viterbi decoding unit 222, and converted by the descrambling unit 224 into a received data signal.

Note that the antennas 202.1 to 202.4 are configured to be shared with the antennas 120.1 to 120.4 of FIG. be able to.

In an arrangement like that of FIG. 2(a), the use of an amount of absorber on the inner surface of the server rack 30 reduces the delay spread, and thus the guard interval required when orthogonal frequency division multiplexing (OFDM) modulation is used. The length of the (periodic prefix period) can be reduced.

Communication in the 60 GHz band is characterized by high transmission loss caused by absorption of oxygen, using low transmit power, and communication is generally short range. The metal structure of the cabinet facilitates the production of multiple delayed reflections, creating channels with frequency selective fading.

In normal operation, the radio channel is essentially static when the front access door is closed, with small channel variations in time caused by temperature variations and material expansion.

Then, the communication capacity is affected by fading depending on the exact location of the antenna of each server 60. However, the exact placement of the elements to avoid fading is not practical to consider at the time of designing the cabinet of the server rack 30 due to the short wavelengths of the millimeter waves, rather the servers 60 within the server rack 30 are not considered. After storing, it is realistic to consider it and adjust adaptively.

Therefore, in the present embodiment, an active material that can be selectively switched between an absorber (absorber) and a reflector (reflector) is used to generate diversity and generate a plurality of communication paths. ..

FIG. 5 is a conceptual diagram for explaining the size of the server rack and the antenna direction of the access point (AP).

In the following, in order to verify the effectiveness of the present embodiment by simulation, the size of the server rack is assumed as shown in FIG.

As illustrated in FIG. 5A, the backplane space (area 54 in FIG. 1) of the server rack 30 is assumed to have a size of 0.35×0.7×2.2 m.

A 10.0 cm thick concrete floor forms part of the bottom tier, while the rack and rear of the cabinet enclosure are made of steel.

It is assumed that the dipole antennas of the transmitter Tx and the receiver Rx are located rearward from the server 60 by a distance of 4.0 cm corresponding to the length of the communication module 94.

Four Rx dipole antennas are mounted on each of the 40 servers 60 in the cabinet, while the four Tx access point (AP) antennas are centrally located.

As shown in FIG. 5B, when simulating the effect on the delay spread when the axis of the dipole antenna is changed with the electromagnetic wave absorber installed, the axis direction of the dipole antenna is set to the x-axis. Direction, y-axis direction, and z-axis direction.

The antenna is not limited to the dipole antenna, and may be, for example, a monopole antenna.

Under the assumption of the size of the server rack as described above, the simulation of communication in the backplane space will be further described below.
[Channel characteristics]
First, but not necessarily, it is assumed that the physical layer (PHY layer) has the same channel bandwidth, modulation and transmission power as the IEEE 802.11ad standard.

In this standard, there are four 2.16 GHz channels that can use the OFDM mode.

A 6.75 Gbps link can be achieved by using 64QAM modulation in OFDM mode with Low Density Parity Check Code (LDPC) (rate=13/16).

The IEEE802.11ad scheme does not include a multiple input multiple output (MIMO) option, but in order to increase communication capacity, the scheme of the present embodiment is revised to include multiple antennas.

The theoretical maximum value of 27 Gbps/rack or 2.7 Gbps/server (time division multiplexing mode: TDD mode) is possible by using a 4×4 spatial multiplexing scheme, although this value is, for example, correlated. Due to the presence of the channel, the handshake and the guard interval, it is actually a lower value.

The above channel characteristics are merely examples, and different characteristics may be assumed depending on the situation.
(Delay spread and transmission loss)
The arrangement of the radio wave absorber affects the delay spread and the propagation characteristics as well as the direction of the dipole antenna.
(Absorber and reflector)
FIG. 6 is a diagram showing an active area arranged in a cabinet of a backplane space of a server rack.

As shown in FIG. 6A, three active regions each having a size of 66×35 cm are provided on both sides of the inner surface of the cabinet.

As shown in FIG. 6B, the bottom portion of the server rack 30 is made of concrete instead of the bottom plate, and the active regions 1 to 3 are provided inside the left side surface from the top plate to the bottom portion. Similarly, the active regions 4 to 6 are provided inside the right side surface.

Note that the number of active regions provided on one side surface is not limited to three, and may be a region divided into a larger number.

Then, each of such active regions selectively functions as a reflector or an absorber by a variable metamaterial that is a variable material constant member as described below. As the variable material constant member, not only a variable metamaterial material as described below is used, but also another member may be used as long as it has a variable dielectric constant/permeability with respect to electromagnetic waves.

That is, the surface structure of a metamaterial capable of changing the transmittance or the reflectance of an electromagnetic wave having a specific frequency according to the applied voltage is disclosed in the following document, for example.

Known Document 1: B. Zhu, K. Chen, N. Jia, L. Sun, et al., "Dynamic Control of Electromagnetic Wave Propagation with the Equivalent Principle Inspired Tunable Metasurface," Nature-Scientific Reports, pp .1-7 , May 2014.
Here, the configuration of the metamaterial disclosed in the known document 1 will be briefly described.

FIG. 7 is a diagram for explaining the configuration of the variable metamaterial described in the known document 1.

In the configuration of the two-dimensional periodic structure metamaterial shown in FIG. 7, the surface of the metamaterial (hereinafter, referred to as “meta surface”) selectively selects an electromagnetic wave of a predetermined frequency according to the applied voltage. It can be reflected or transmitted.

Each of the two-dimensional periodic structures consists of a parallel LC resonant tank and a small loop of metal to provide the required electromagnetic reactance. The varactor (variable capacitance capacitor) is integrated in the unit cell so that the surface impedance of the meta surface of the metamaterial can be dynamically adjusted.

What is described as a high resistance wire is a high resistance bias wire for supplying a bias voltage to the varactor. Such a structure is basically transparent to high frequency electromagnetic waves.

Further, the constant-capacity capacitor arranged in parallel with the varactor has a function of adjusting the resonance frequency of the loop including the varactor and of decoupling the DC bias to the varactor.

In the configuration as shown in FIG. 7, the surface electric reactance is expressed by the following equation.

Where Z _LC is the impedance of one LC resonant tank, Z _d is the impedance of the varactor, L is the value of the parallel inductance, and 2g and 2b are the width and height of the unit cell. Yes, ω is each frequency.

On the other hand, the magnetic impedance of the surface is expressed by the following formula.

Here, μ is the effective magnetic permeability of the meta surface, and a represents the size of the loop in the meta surface structure in FIG. 7.

By properly designing the LC resonant tank and the metal loop, the metasurface should have the electromagnetic surface impedance required for transmission or reflection of electromagnetic waves of any frequency. You can Then, by controlling the varactor capacity, the resonance frequency of the LC resonance tank and the metal loop can be dynamically adjusted.

By adopting a structure in which an absorber is provided on the back surface of the surface structure made of a metamaterial as shown in FIG. 7, each active region functions as an absorber in a high transmittance state and in a high reflectance state. It will function as a reflector.

By using such materials, the reflectivity and thus the properties of multipath transmission can be adaptively controlled.

Therefore, some of the active regions shown in FIG. 6 are selectively set to the inactive state (state of the absorber) under the control of the control unit (not shown), and the remaining regions are activated to the active state (reflection). Body condition).

The function of the control unit that performs such control may be realized by providing an arithmetic processing unit inside the L2/L3 switch 62, or by implementing the arithmetic processing unit outside the server rack 30. A computer for performing such control may be separately provided in common for the server rack 30 of FIG.

Hereinafter, in the case where the material arranged in each active region is capable of switching between the function as an absorber and the function as a reflector in this way, a metric for performing the switching will be examined by simulation.

In this simulation, the characteristic is modeled as being 1 mm thick steel when the active region is the reflector in the active state, and the characteristic has a reflection coefficient of -23 dB when the active region is inactive. Modeled as a dissipative/lossy dielectric (absorber) with.

The front panel is made of steel mesh with holes sized to allow heat to escape from the cabinet as described above, but EM radiation is reflected backwards for security reasons. ..

Here, it is assumed that an absorber is placed on a part of the front surface which covers 15% of the surface area in order to reduce the delay spread. However, the area of the absorbent material on the front side may be enlarged or reduced, and is not limited to this area.

The effect of the active area can be estimated from the signal direction of arrival (DoA) plot and is evaluated using wireless ray tracing software.

FIG. 8 is a diagram showing a simulation result of such wireless ray tracing.

FIG. 8A shows a case where the active region {2, 6} is active ({A, B} represents a position in the active state (reflection state) in the active region of FIG. 6A). Rx position (1,5) ((X,Y) is the XZ coordinate when the unit length is 4.4 cm. For example, (1,5) is 4.4 cm or 22.0 cm from the lower left. ,) at the antenna position (hereinafter, PA). The upper row shows the received power of the incoming radio wave with respect to the angle in the φ direction, and the lower row shows the received power of the incoming radio wave with respect to the angle in the θ direction.

Here, in the XYZ axes shown in FIG. 5B, θ is an angle with respect to the Z axis (parallel to the cabinet depth direction), and φ is within the xy plane with respect to the X axis (parallel to the cabinet width direction). There is an angle.

FIG. 8B shows the arrival direction at the position (1, 5) of Rx (4.4 cm from the lower left, 22.0 cm) when the active region {3, 4} is active. The upper row shows the received power of the incoming radio wave with respect to the angle in the φ direction, and the lower row shows the received power of the incoming radio wave with respect to the angle in the θ direction.

The DoA at arbitrarily chosen receiver Rx coordinates (1,5) are shown in positions {2,6} and {3, as shown in FIGS. 8(a) and 8(b), respectively. 4} are plotted against the two active regions.

In both graphs, it can be seen that the dominant LOS (Line Of Sight) line is present, but the reflected multipath comes from different directions.

FIG. 9 is a plot of the corresponding wireless ray tracing software.

In the figure, FIG. 9A shows the paths of the LOS and the reflected wave when the active region on the right side of the bottom surface of the side surface is in the active state (reflector), and FIG. The path of LOS and a reflected wave when an active region is inactive (absorber) is shown.

At the position coordinate (1, 5) (indicated by PA in the figure), the LOS and the strongest reflection path are shown from the center TX1 to Rx.

It can be seen that the route of the radio wave received by the server 60 changes greatly depending on whether the active region is in the active state or the inactive state.
[Software simulation]
As will be described later, the control unit as described above changes the optimum active area for each server rack while changing the set of active statuses of active areas, according to the server placement, environmental conditions, etc. at that time. A set of active states is searched and set to such an optimum set of active states. Such searching and setting are performed, for example, in the data center during the calibration work after the installation work of the server rack and the server is completed. It should be noted that due to changes in the conditions inside the server rack (changes in temperature, changes in the position of servers in the rack due to maintenance work, changes in the placement of objects in the server rack, etc.), for example, at regular intervals. Alternatively, the calibration by the search and setting as described above may be executed.

In the following, in order to evaluate the effect of such search and setting on the communication capacity of the server, the simulation capacity setup of the ray trace data and the calculation result of the communication capacity by the post-processing will be described.
(Ray tracing model)
FIG. 10 is a plan view for explaining the arrangement of Rx antennas and Tx antennas used in the simulation.

The radio channels shall be observed at 49 vertical and 15 horizontal receiver (Rx) positions at regular intervals of 4.4 cm, and the simulation shall give a total of 735 observation locations. ..

The servers start at the bottom and are numbered Sv1 to Sv40.

Forty servers are stacked in consecutive rows with a spacing of 4.4 cm or 8.8λ. Also, extra row spacing is inserted between every four racks supporting ten servers.

The linear array of four receiving (Rx) antennas provided in each server 60 has an element spacing of 8.8λ and is arranged in a row in the lateral direction.

Further, in the transmitting (Tx) antenna provided in the L2/L3 switch 62, antenna elements are arranged at respective apexes of a rectangle having a longitudinal direction of 4.4λ and a lateral direction of 8.8λ.

In addition, in the simulation, a total of 23 different combinations such as the following are considered as the “set of active states of the active region”.

i) Sets 1-15 contain 15 permutations that specify two active regions from six positions.

ii) Set 16 has all active regions active.

iii) Set 17 has none of the active regions inactive.

iv) Sets 18-23 are where one region is active.

More generally, there are a total of 2n active regions on both sides of the cabinet, of which 0, 1,... M regions and 2n are all in the active state (reflector state). The number N is as follows.

FIG. 11 is a diagram showing simulation parameters.

The channel impulse response is also calculated by simulation.
(Send unique mode)
The 4x4 channel of each subcarrier is decomposed into two unitary matrices V and U using singular value decomposition (SVD).

Here, λn is the nth eigenvalue, and H is a channel matrix of 4×4 channels.

The first eigenvalue depends strongly on the total received power and the LOS signal.

Its value is almost the same with or without an active reflector.

The reflected signal has a lower power and makes only a small contribution to the first eigenvalue. As the number of individual reflections increases, the relative value of the second eigenvalue generally increases.
(Communication capacity)
The communication capacity of each link depends on the spatial correlation of the signals incident on the Rx array element.
Therefore, the range of communication capacities is location-dependent, provided that some servers have relatively higher values and others have lower values.

At high SNR, the communication capacity is given by equation (4).

Here, λk is the k-th eigenvalue, where r _H is the channel rank or the number of non-zero singular values. σ _n ² is the noise variance, and P is the magnitude of the substream transmission power vector.

The optimized distribution of power in each eigenmode depends on the SNR of the substream and is determined by the Water-filling algorithm.

Here, the water injection theorem algorithm is disclosed in the following documents.

Known Document 2: C. Oestges and B. Clerckx, “MIMO Wireless Communications: From Real-World Propagation to Space-Time Code Design”, Academic Press (Elsevier), Oxford, UK, April 2007.
The received SNR is set to 20 dB based on the link budget estimate.

Here, "link budget" means gain factors (gain due to antenna, amplifier, coding, etc.) existing in a path (link) between a transmission end and a reception end of a communication system and loss factors (cable, free space). , Fading, loss due to thermal noise, etc.) are all added/subtracted in dB value to calculate the balance (allowable propagation loss) of the entire link. From this calculation, the arrival distance of the radio wave from the transmission source to the reception end can be known.

The capacity is summed for the top four active substreams for each subcarrier.

For each of the 40 servers, the capacity is recorded for each subcarrier and the process is repeated, shifting the Rx array at positions 1-12.

The simulation is repeated for each of the 23 active area sets.

FIG. 12 is a diagram showing a distribution of communication capacities when antennas are arranged as shown in FIG.

Depending on the set of active areas, considerable variability exists for servers SV1-SV15 and servers SV30-40 on the bottom and top side of the cabinet.

The capacity is more correlated for the central servers SV15-25 and is taken into account by the relatively stronger Lician component of the LOS, which dominates the contribution of multipath transmission.

For example, the communication capacity for the server 18 is increased from 14.4 bits/s/Hz using the active area set 4 to 23.2 bits/s/Hz by using the active area set 21.

FIG. 13 is a diagram showing a cumulative distribution function (CDF) of the communication capacity for each of the sets of active areas.

At the median (50% CDF level), an SNR of 20 dB gives the best set (set 13) a higher 2.32 bits/transmission capacity than for the active region set 12.

Therefore, the set 13 of active regions is considered to be the optimum set for a particularly accurate arrangement of Tx and Rx in the simulation.

However, at 60 GHz, an antenna displacement of just 0.125 mm corresponds to a phase difference of 9.0 degrees.

Therefore, such an exact placement is not practical as a single placement in the manufacturing or installation stage, and in order to find the best set, the controller activates it at least once the installation is complete. It is necessary to control the switching of active and inactive regions.

Further, it is desirable to dynamically control switching between the active state and the inactive state of the active region in accordance with the operating time of the server, for example, at regular intervals. This is because a small amount of movement of the active region, introduction of foreign matter, thermal expansion of the metal, etc. all contribute to the optimum set change.

The active area can be adaptively set and optimized for the traffic demand of each server.
(Condition number (number of conditions))
In the ideal state, the eigenvalue of the substream has the same value for maximizing the communication capacity.

The condition number (CN: Condition Number) is the ratio of the largest eigenvalue to the smallest eigenvalue and is a useful indicator of how well the channel matrix is conditioned.

It is represented by the following equation (5) in each subcarrier i.

When CN is less than about 10 dB, the channel matrix is considered in good condition, while detecting this value is more sensitive to errors in channel estimation.

FIG. 14 is a diagram showing CDF simulation results with respect to the condition number for each of the sets of active regions.

The CDF results, as shown in FIG. 14, show that 59% of the channels have a condition number less than 10 dB.

As the number of conditions increases, so does the SNR required to maintain a given level of performance.

The calculation of CN can be performed with almost no increase in calculation complexity in practical systems where CSI has already been calculated and CN value is used to control the switching of the active area.

For example, if CN is below the 10 dB threshold, the active region set is considered good, and if that condition is not met, an alternative configuration can be tried.
[Control method of control unit]
(Control so that the median value of communication capacity is the highest)
FIG. 15 is a flowchart showing a first procedure of the control procedure of the control unit, which searches for an optimum activation group of the active area for a certain server rack.

The control unit performs the simulation as shown in FIG. 15 for each server rack, and actually controls each server rack to switch the active state and the inactive state of the active area.

In the following, it is assumed that the number of sets of active states of the active region is N _S.

With reference to FIG. 15, when the process is started, the arithmetic unit of the control unit first performs the following initialization process (S100).

i) With respect to the set of active states (1 to N _S ) of the active areas, the active areas to be activated are specified from the information stored in the storage device.

ii) The variable best_set representing the optimum set is set to the initial value 1.

iii) A variable s designating a set of active states of active regions to be processed is set to an initial value 1.

iv) The communication capacity evaluation threshold value CAP _thr is set to an initial value of 0 bits/s/Hz.

Subsequently, the arithmetic unit of the control unit evaluates the channel matrix H from the received data according to a known method for the number of reception antennas N _RX and the channels of N _RX ×N _TX of the transmission antennas N _TX (S102).

Here, the evaluation of the channel matrix H is performed for all the sets s of active states of the active region, the server ν, and the subcarrier i.

Next, the arithmetic unit of the control unit decomposes the evaluated channel matrix H into singular values in the same manner as in the equation (3) (S104).

The arithmetic unit of the control unit accumulates the communication capacities calculated in the same manner as in Expression (4) for each server (ν) and each subcarrier (i), and calculates the cumulative distribution function (CDF) value of the communication capacities. The CDF Cap(s) is calculated (S106).

Subsequently, the arithmetic unit of the control unit, CDF median Cap _50% of Cap (s) and (s), compares the evaluation threshold CAP _thr, if towards the median Cap _50% (s) is greater (Y in S108), the variable best_set is set to the current value of the variable s, the evaluation threshold value CAP _thr is set to the median value Cap _50% (s) (S110), and the process proceeds to step S112.

On the other hand, when the median value Cap _50% (s) is equal to or _smaller than the evaluation threshold value CAP _thr (N in S108), the process proceeds to step S112.

In step S112, the arithmetic unit of the control unit determines whether the number s of active states of the active region has reached the maximum number N _s , and when the number has not reached the maximum number (N in S112), the variable s is set. It is incremented by 1 (S114) and the process is returned to step S106.

On the other hand, the arithmetic unit of the control unit ends the process when the maximum number is reached (Y in S112).

By executing such processing, the variable best_set is set to the optimum active area set s based on the highest median communication capacity.
(Control so that the number of communication capacity conditions is the lowest)
FIG. 16 is a flowchart showing a second procedure of the control procedure of the control unit, which searches for an optimum activation group of the active area for a certain server rack.

The control unit performs a simulation as shown in FIG. 16 for each server rack, and actually performs control for switching the active state and the inactive state of the active area for each server rack.

In the following, the number of active state pairs in the active region is N _S.

With reference to FIG. 16, when the process is started, the arithmetic unit of the control unit first performs the following initialization process (S200).

i) With respect to the set of active states (1 to N _S ) of the active areas, the active areas to be activated are specified from the information stored in the storage device.

ii) The variable best_set representing the optimum set is set to the initial value 1.

iii) A variable s designating a set of active states of active regions to be processed is set to an initial value 1.

iv) The condition number evaluation threshold CN _thr is set to an initial value of 1000 dB.

Next, the arithmetic unit of the control unit evaluates the channel matrix H from the received data according to a known method for the number of reception antennas N _RX and the channels of N _RX ×N _TX of the transmission antennas N _TX (S202).

Here, the evaluation of the channel matrix H is performed for all the sets s of active states of the active region, the server ν, and the subcarrier i.

Next, the arithmetic unit of the control unit performs singular value decomposition on the evaluated channel matrix H in the same manner as in Expression (3) (S204).

The arithmetic unit of the control unit accumulates the condition number CN for each server (ν) and each subcarrier (i) in the same manner as the equation (5), and calculates the cumulative distribution function (CDF) value CDF CN of the condition number CN. (S) is calculated (S206).

Subsequently, the arithmetic unit of the control unit, the median CN _50% of CDF CN (s) and (s), evaluated by comparing the threshold value CN _thr, if towards the median CN _50% (s) is less (Y in S208), the variable best_set is set to the current value of the variable s, the evaluation threshold CN _thr is set to the median value CN _50% (s) (S210), and the process proceeds to step S212.

On the other hand, when the median value CN _50% (s) is equal to or larger than the evaluation threshold value CN _thr (N in S208), the process proceeds to step S212.

In step S212, the arithmetic unit of the control unit determines whether the number s of active states of the active regions has reached the maximum number N _s , and when the number has not reached the maximum number (N in S212), the variable s is set to It is incremented by 1 (S214) and the process is returned to step S206.

On the other hand, if the arithmetic unit of the control unit has reached the maximum number (Y in S212), the processing ends.

By executing such a process, the variable best_set is set to the optimum active region set s based on the median of the condition numbers being the lowest.

In FIGS. 15 and 16, the optimum active area set s is set based on the median of the communication capacity and the median of the condition number (condition number). Based on other conditions such as the lowest value, the lowest communication capacity, the highest median received power, and the like, the optimum active region set s is determined. You may search and set.

Therefore, in the present embodiment, as described above, the value representative of the communication state, which is the reference when searching for the set s of the active states of the optimum active region, is referred to as a “communication metric”. ..

As described above, with the server rack, the server, and the communication module according to the present embodiment, it is possible to efficiently reduce the influence of delayed waves even in a metal enclosed space in the server rack, and prevent deterioration of communication quality. it can. Most of the cables in the server rack can be removed, and as a result, the cooling efficiency of the server in the server rack can be improved and the power consumption for cooling can be reduced.

In addition, when communicating in a server rack with electromagnetic waves having a short wavelength such as millimeter waves, even if a large number of servers are actually stored in the server rack, the position of the server and antenna during storage may be subtle. The above effects can be achieved by calibrating to absorb the effects of misalignment, temperature changes, server misalignment during maintenance, and the effects of newly installed items in the server rack. ..

The embodiment disclosed this time is an exemplification of a configuration for specifically carrying out the present invention, and does not limit the technical scope of the present invention. The technical scope of the present invention is shown not by the description of the embodiments but by the scope of the claims, and includes modifications within the literal scope of the claims and the scope of equivalent meaning. Is intended.

1-6 active area, 30 server rack, 40 bottom plate, 42 top plate, 44, 46 side plate, 48 back plate, 50 front door, 52 opening, 60 server, 62 L2/L3 switch, 92 USB terminal, 94 communication module.

Claims (5)

A housing structure for an electronic device having a housing that defines a space for housing a plurality of electronic devices,
A housing having a front half having an opening and a closed rear half;
The internal space of the housing is divided into the front half section and the latter half section by accommodating a plurality of electronic devices therein, and the rear surface of the plurality of electronic apparatuses is the latter half section. Arranged to face the side, each of the plurality of electronic devices has a communication module on the back side for transmitting and receiving radio signals of a predetermined frequency,
An active region provided in each of a plurality of regions provided on the inner surface of the housing;
Each said active region has an absorbent member provided on the rear side of the the installed that variable material constant member to direct the front inside the housing variable material stereotactic number member,
The surface of each of the plurality of variable material constant members on the inner side of the housing can switch the function as a transmissive body or a reflective body with respect to a radio signal of the predetermined frequency under the control of the control unit. Yes,
In the calibration period, the control unit adaptively switches the surface states of the plurality of variable material constant members based on a communication metric when switching to a transmissive body or a reflective body ,
The communication metric is a median value of communication capacities of communication of the plurality of electronic devices,
The control unit sets the surface state of the plurality of variable material constant members so that the median value of the communication capacity is maximized . A housing structure for an electronic device having a housing that defines a space for housing a plurality of electronic devices,
A housing having a front half having an opening and a closed rear half;
The internal space of the housing is divided into the front half section and the latter half section by accommodating a plurality of electronic devices therein, and the rear surface of the plurality of electronic apparatuses is the latter half section. Arranged to face the side, each of the plurality of electronic devices has a communication module on the back side for transmitting and receiving radio signals of a predetermined frequency,
An active region provided in each of a plurality of regions provided on the inner surface of the housing;
Each of the active regions has a variable material constant member installed so that its front surface faces the inner surface side of the housing, and an absorbing member provided on the back surface side of the variable material localization member.
The surface of each of the plurality of variable material constant members on the inner side of the housing can switch the function as a transmissive body or a reflective body with respect to a radio signal of the predetermined frequency under the control of the control unit. Yes,
In the calibration period, the control unit adaptively switches the surface states of the plurality of variable material constant members based on a communication metric when switching to a transmissive body or a reflective body,
The communication metric is a median of condition numbers of communication of the plurality of electronic devices,
The control unit sets the surface condition of the plurality of variable material constant members so that the median value of the condition numbers is minimized . For each of the plurality of variable material constant members, a set of a plurality of states to be a transmitter or a reflector is predetermined,
The control unit calculates the communication metric for each of the plurality of state sets, and sets the surface state of the plurality of variable material constant members to an optimum state of the plurality of state sets. The housing structure for an electronic device according to claim 1 or 2 . The predetermined frequency is a millimeter wave band,
The electronic device is a server, receiving structure of the electronic device according to any one of claims 1-3. The housing has a rectangular parallelepiped shape including a top plate, a bottom plate, two side plates, a back plate, and a front door provided with an opening,
The plurality of servers are housed and fixed inside the housing so as to be separated from the back plate by a certain distance in an internal space defined by the housing, whereby the back plate, the bottom plate, and the top plate, The latter half is defined by the two side plates and the back surfaces of the plurality of servers,
The housing structure for an electronic device according to claim 4 , wherein the variable material constant members are respectively installed on the inner surfaces of the two side plates.